Review of Image Search Engines
This is the fourth JISC Digital Media review of image search engines In addition to this review, we have published reviews of moving image and audio search engines.
February 2008
- Introduction
- Image search engines and their limitations
- Evaluation criteria and test search terms
- Major changes since previous review
- Image search engine reviews
- Meta- searching and 'springboard' searching
- Content-based image retrieval
- Conclusion
1. Introduction
This is the fourth JISC Digital Media review of image search engines (earlier reviews have been archived to enable comparison: 2006, 2004 and 2003). Readers should note that in addition to this review, JISC Digital Media has published a Review of Moving Image Search Engines and Review of Audio Search Engines.
Please note that the information presented here is based on research conducted in January 2008. If you're reading this some time after that date, you are advised to check the sites carefully and consider asking similar questions to those JISC Digital Media has included in its evaluation criteria (see 3 below).
2. Image search engines and their limitations
By "image search engines", we mean Web-based services that collect and index images from other sites on the Internet. Image searching is offered by general search engines, such as Google or Yahoo, and by some specialised search engines – services devoted to the searching of images or multimedia. In addition, there are a number of "meta-search engines", which pass on search queries to more than one search engine and then bring you back the results. We will look at examples of all three kinds of engine below. Section 4 also considers, briefly, some attempts to create content-based search engines, which 'index' images according to visual characteristics such as the distribution of colour or shapes. To date, such engines have been experimental and based on individual collections rather than the Web as a whole, but this technology will be used more in the future and is now being used in a small way by some of the main image search engines.
The reference to visual indexing in the previous paragraph highlights an important "deficiency" in current image search technology:
- Image search engines have to rely on text for their indexing. As a consequence of this, the quality of an engine's image results depends on the quality of the textual information that surrounds or is associated with the images (e.g. filename, nearby text, page title, or 'alt' picture tags within the HTML code). Although search engines are becoming increasingly sophisticated in processing these textual clues, their image indexing 'algorithms' are essentially guessing at the content of images. They are no competition for a good human eye. This explains why the results from such engines will often appear poor or mismatched. It also suggests that for some kinds of image searching it might be better to turn to an image collection that has been indexed by people rather than machines!
There are a few other important limitations of image search engines that you should be aware of:
- Image search engines cannot index all the images on the Web. Although most aim to be as comprehensive as possible, image search engines exclude a vast number of online images. Those managing Web sites or online collections can choose to deliberately shut out the search engine "spiders" (indexing tools). Or they may be using forms of technology that do not allow indexing to take place (e.g. password protected databases or dynamically generated Web pages). Lots of collection owners want their resources to be found by search engines and are finding ways of overcoming these technological limitations, but it's still the case that many of the very best images on the Web are tucked away beyond the reach of search engines in a region that some term the "invisible Web".
- Image search engines will seldom let you see all the images they have! While the larger engines claim to have billions of images, almost all of them cap their results at a thousand or less. This is a lot of images, but given their mixed results, it can leave you wanting. So when using and evaluating image search engines, the quality of the first few pages of results is much more significant than the quantity they report.
- Image search engines are never completely up-to-date. Image search engines never search across the whole Web whenever you enter a search query – this is impossible for current technology. What they do is compare your text with the huge indexes they've prepared in advance and then provide you with pre-generated thumbnails and links to the relevant Web sites. It can take many weeks to index the whole Web, during which some sites will have changed or been removed. This is why you sometimes click on a thumbnail and encounter a "dead" link or search in vain to find the image on the page! The indexing delay also means that it may take a long while for new images to find their way into a search engine index. The better engines now typically prioritise or target news and celebrity sources, so it's easier than it used to be to find images for current events, but it may be better to try other sources for these kinds of images.
While image search engines certainly have an important role to play in finding images, it should be clear from the points above that they are by no means a "one-stop-shop". In addition to the limitations imposed by text-indexing, incomplete coverage, partial results and lack of currency, they cannot make adequate assessments of: (1) the aesthetic quality of images they index; (2) copyright restrictions on their reuse; or (3) an image's suitability for your particular purpose. These are key things you will need to consider when deciding whether to use an image search engine and what to make of their results.
So where else could you go? In addition to this review, JISC Digital Media is preparing a series of advice documents on finding and using multimedia resources, which provide pointers to other sources of images, video and sound.
3. Evaluation criteria and test search terms
To assist in the evaluation and comparison of the search engines in this review, JISC Digital Media asked several questions (see Evaluation Criteria table below) and used the same set of search terms to test each engine (see Search Terms table). We have provided some phrases to indicate our overall evaluations, which are defined in the Overall Evaluation table (below). Readers should be aware, however, that an exercise of this nature is necessarily subjective: one reviewer's judgement of "quality", for example, is very likely to differ from another's. The reviews we make here are intended to provide the reader with a resource to assist in choosing an appropriate search service, but they make no claim to scientific accuracy. We summarise the results of our tests in this review. It any reader would like further information about the tests or our findings, please contact JISC Digital Media's helpdesk.
Readers should also be aware that this review was conducted in late January 2008. If you're reading this some time after that date, you may wish to make your own evaluations – perhaps using some of the questions in the criteria below.
Evaluation CriteriaIn reviewing image search engines, JISC Digital Media asked the following questions: |
|
|---|---|
| Scope |
|
| Search Options |
|
| Performance |
|
| Presentation |
|
| Support |
|
Test Search TermsJISC Digital Media tested the image search engines using the following terms: |
||
|---|---|---|
| Category | Terms and phrases | Rationale |
| General search terms |
|
General, relatively unambiguous, terms |
| Specific search terms |
|
Scientific name to test the indexing and an unambiguous historic figure to test scope and relevancy |
| Ambiguous search terms |
|
To test the weighting of the indexes and their handling of phrases |
| Specific images |
|
Two unique filenames to test the comprehensiveness and currency of coverage: the JISC Digital Media image has been online for several years; the BBC image was published a week before the review was conducted. The Picasso image is included to see how a famous in-copyright image is included and presented |
Overall evaluationIn the reviews that follow, JISC Digital Media has used the phrases "Poor", "Average", "Good" and "Very good" to provide an overall evaluation for each of the criteria listed above (apart from scope). These are based on a comparison of all the engines being reviewed and mean the following: |
|
|---|---|
| Very good | Among the best for this criteria |
| Good | Better than average, but not outstanding, for this criteria |
| Average | Mid-range for this criteria |
| Poor | Below average for this criteria |
4. Major changes since our previous review
At the time of our last review (mid-2006), Google and Yahoo were the major image search providers among the general search engines, with very strong competition from the specialist search service Picsearch. Ask.com was a promising new entrant and there were hints that Microsoft was preparing to launch its own service (at the time of the review it was using Picsearch). Microsoft did indeed launch a service, for images and for the wider Web, which it has called Live Search. At the time of this current review (early 2008) the phrase "The Big Three" is often used as a shorthand for describing Google, Yahoo and Live Search. Although estimates vary, it seems likely that Google deals with at least half of all searching done on the Web, with Yahoo and Live Search taking up a sizable chunk of the rest. Image searching probably follows a similar pattern. As this review was completed there was talk of a reconfiguration among "the big three", which might see Microsoft and Yahoo combining to take on the might of Google.
While The Big Three undoubtedly hold the lion's share of images and image searches (Google, Yahoo and Livesearch, in this order), this does not necessarily mean that they are "The Best Three" image search engines. As this review shows, other services (particularly Exalead and Picsearch) may offer some distinct advantages over some among the Three. Exalead is a strong new entrant in this review. Another newcomer is Pixsy.
In terms of the size of their image indexes, Google has pulled well ahead of the competition and now holds a vast index of many billions of images. Comparing results from our 2006 and 2008 tests Google seems to have multiplied its index by 10 or 20-fold! Yahoo has also grown (reporting 2 or 3 times as many results) and others have increased more modestly (Picsearch, for example, seems to have grown by about a third). However it's important to stress that it's the quality of the first few pages of images that really counts, since only Picsearch will let you see all its results.
There have also been some changes in the way we present our review. In previous reviews we have separated out general search engines, specialist image search engines, and engines drawing their results from other search engines. As the main specialist engines now also offer general web searching, we've decided to put all the engines into a single alphabetical listing - including those who draw on others for their results (these have briefer summaries). However the "meta-search" engines, which simply pass on a search query to one or more search engines, continue to be dealt with separately in a later section of this review (section 6).
We have removed two search engines that featured in our previous review: Cydral and A9. Cydral no longer has an English interface and A9 no longer offers public searching, concentrating instead on selling its search technology. We have also excluded Objects Search, which sometimes appears in lists of image search engines. This is because it frequently includes adult images among its results without providing a means of filtering them out. We note that the results Objects Search provides are also very small and of mixed quality (e.g. our search on "Queen" produced about 400 results).
5. Image search engines reviews
Reviews are presented here in alphabetical order. We have reviewed every significant English-language image search engine we are aware of, although have provided briefer descriptions for those that rely on other engines for their results (e.g. AlltheWeb, which uses Yahoo's index).
We also reviewed Alexa's image search engine, but at the time of publication Alexa is no longer providing one.
AlltheWeb (uses Yahoo)
http://www.alltheweb.com/?cat=img
AlltheWeb is owned by Yahoo and draws on it for its image searching. It offers slightly different advanced search options to the main Yahoo image search, provides fewer filters after searching, and dispenses altogether with the split-screen or framed results commonly employed by other engines. Instead: clicking on a thumbnail opens the image fullsize on a new page, while clicking on the link under the thumbnail takes the user directly to the original website with the image in situ. Copyright issues are not drawn to the user's attention, but are mentioned in an FAQ. The most significant difference between Yahoo and AlltheWeb is that the latter does not give as much prominence as Yahoo to images from Flickr (see below).
AltaVista (uses Yahoo)
http://www.altavista.com/image/
Like AlltheWeb, AltaVista is now owned by Yahoo and draws on its index although the search options and presentation of results is different. Notably, like AlltheWeb, the results lack the Flickr images that have come to dominate Yahoo's results (see below). Altavista enables its users to limit results, before or after a search, by colour, size or a source type (e.g. 'news' or 'web'). Its results page provides a thumbnail, filename, pixel dimensions and file size, with a link to 'more info'. This link provides some further information about an image, including colour information, source URL address and, usefully, an 'abstract' (descriptive text excerpted from the source Web page). AltaVista also usefully lists other pages on which it thinks the same image may have been used. If, at the results page, the user clicks on the thumbnail rather than the 'more info' link, they are taken directly to the source Web page of the image. No copyright warnings are given on AltaVista's results pages or within its online help pages.
AOL (uses Google)
http://search.aol.com/aol/imagehome
http://search.aol.co.uk/aol/imagehome
AOL draws on Google for its results. Its results are slightly different from Google's, but it is not clear if this is due to the freshness of its Google results or the addition of other image results. AOL.com (the US site) has an advanced search and results screen that are almost identical to Google's. However, when you click on a thumbnail image you are given less information, with no copyright warning, and the original Web source fully framed within the AOL page. AOL.co.uk (the UK site) offers no advanced search or user preferences (adult material is filtered out by default). But it is similar to the US site in its presentation of results.
Ask
http://uk.ask.com/ (Click on the image icon to search for images)
Ask was a new entrant in JISC Digital Media's last search engine review, having arrived in January 2006. Prior to this, it had relied on Picsearch for its images. Ask has improved in its presentation and performance since the last review, but is still limited by its small number of results and lack of advanced search options. However, it does accompany its results with some very useful additional information, links, and services (e.g. its "MyStuff" bookmarking). Ask currently provides image services to Lycos US (below) although this looks likely to change during 2008.
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Big and broad | Ask does not indicate the size of its index. In our tests it tended to report results close in size to Exalead and Picsearch, but significantly less than LiveSearch, Yahoo and Google. But note that you can only actually view up to 320 results. |
| Search Options | Good | Ask does not provide an advanced image search or any initial search limitation (e.g. by size or colour). However it does enable you to filter after a search by size, colour, and file type. It also makes helpful suggestions as you type in your search, prompts you if it thinks you might have misspelled something, and suggests some further search terms that may be 'related' or enable you to 'expand' or 'narrow' your search. Adult material is filtered out by default. |
| Performance | Good | Results are fairly quick and generally appear relevant and of good quality. We did not observe any missing images, dead links, or duplication among the results. Despite the default filtering, the occasional adult image was spotted. |
| Presentation | Good | Results are provided as thumbnails (12 at a time), along with pixel dimensions, file size and source Web site. There are also links to related web content (including news and video) and some relevant adverts. Once a thumbnail is clicked, a split screen view provides further information about the image at the top (filename, URLs for the image and source pages, format, and a copyright warning). The original page is presented below. Ask has a "MyStuff" facility that enables users to bookmark images for later reference. |
| Support | Average | There is no immediately obvious help information. Ask's 'About' page at the bottom links to its 'Help Centre', but this does not provide specific help on image searching. It does, however, provide a contact form where users can request specific help. |
| JISC Digital Media's test words | Good | Ask did not find the JISC Digital Media or BBC image, but found a reasonably good range of results for the other terms – although the cap of 320 results means that the choice can sometimes be quite limited. |
Ditto (uses Picsearch)
Ditto draws on Picsearch for its image searching, but provides no more than 228 results (19 screens of 12 images) and offers no advanced search options. Results are displayed as thumbnails which link directly to the original source page, rather than the split-screen or framed presentations more commonly used by image search engines. The only other information Ditto provides about an image is its file size. There is a notice at the bottom of each page advising users that they must obtain appropriate permission to use the image. Ditto's copyright warning and its avoidance of the split-screen may be a consequence of a 2002 copyright suit which was bought against Arriba (Ditto's previous identity) for the way it presented other people's images within framed Web pages.
Exalead
http://www.exalead.co.uk/image/
Exalead is a significant new entrant on our list. Provided by a French company, it was offering English language searching by 2004 and image searching by 2006, but has grown in prominence and use during 2006 and 2007. A new version of Exalead's image search engine was launched in April 2007, which added more sophisticated search filtering, including face recognition technology. We were impressed by the quality of some of its results and, particularly, by its filters and advanced search options.
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Big and broad | Exalead does not indicate the size of its index. In our tests it reported results close in size to Ask and Picsearch, but significantly less than LiveSearch, Yahoo and Google. But note that you can only actually view up to 1000 results. |
| Search Options | Very good | Exalead provides simple searching and some very good advanced options, which will apply Boolean operators, limit by Web site, and filter by size (including the ability to specify particular pixel dimensions). Once results are obtained, they can be filtered by size, colour and for "faces" (images which have been tagged as likely to contain faces by content-recognition software). Note that Exalead offers some "more choices", which provide further search limiting (e.g. layout and file type) and, very usefully, give an indication of what numbers or percentages of results relate to each image characteristic. As well as being able to limit your results to one of these characteristics, Exalead enables you to exclude images by particular characteristics (e.g. you can exclude all small images). Adult content filters are available within Exalead's advanced search and within the user's 'preferences' (where the default is moderate filtering). |
| Performance | Very good | Results appear quickly and are of generally good relevance and quality. We did not observe any missing images, dead links, or adult content, but saw some duplication (the same image on different sites). |
| Presentation | Very good | The results are provided quickly with 21 thumbnails per page. It is possible to set these thumbnails against a black background if preferred. Reasonably high numbers of results were report, although it is not possible to view more than 1000. Thumbnails are accompanied by a title or filename, pixel dimensions, file size and file type, and source Web site. Clicking on the thumbnail will bring up a new screen with this same information at the top and the source page framed below it. It is also possible to click through to the previous and next results, with their source pages displaying in the frame below. From the initial results screens you can alternatively 'view image alone' which will open up an image by itself on a page. Or you can 'Add to shortcuts' which creates a link to the image from a personalised homepage. There are no obvious references on the main pages to potential copyright restrictions. |
| Support | Very good | Exalead has good help information, which does include some mention of copyright. It also offers a feedback prompt for users at the bottom of each results page ('dissatisfied with the results? Help us improve'). |
| JISC Digital Media's test words | Very good | Exalead did not find the JISC Digital Media image when we searched by filename, but a search limited to the JISC Digital Media Web site did bring it up among the results. It also indexes the BBC news pages, but did not seem to have yet indexed our test image. Results for JISC Digital Media's other searches were generally good and were drawn from a diverse range of sites. On the "Homer" search, the face recognition software ignored the Simpsons cartoon faces but managed to pick up some sculptures of the classical poet. |
Google Image Search
Google is the largest and most popular image search engine. Our tests suggest that it has vastly expanded its image index over the past couple of years, although its continued cap at 1000 images means that users cannot fully explore its findings. Given this limitation, it is important that the most suitable images are shunted to the beginning of the results and Google seem to do this fairly well, delivering consistently good results. Since the last review, Google has made a few changes to its search options and display of results – most notably enabling users to concentrate their searching on news content and on images containing faces.
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Huge and broad | Google does not indicate the size of its image index. Judging by its reported number of results, Google's index is enormous: many times that of Yahoo and LiveSearch, who offer the next biggest results. But note that you can only actually view up to 1000 results. |
| Search Options | Very good | Google offers both simple and advanced searching. Its advanced search enables Boolean and phrase searching and can limit the results by size or colour or to a particular Web site or Web domain. The advanced search also enables you to limit your results to news content and to 'faces' (like Exalead this utilizes content-recognition technology). Google's adult filter can be changed for a particular search (it is normally set in 'preferences' where it defaults to moderate filtering). Once a search has been conducted, it is possible to filter the results by size, or to click on the 'advanced search' option to further refine the search. |
| Performance | Very good | Results are returned very quickly with very high numbers reported (although only 1000 can be viewed). We found very good relevancy and quality among the initial results, with the pages becoming a bit more mixed as later pages were viewed (this suggests that it has a strong sorting algorithm). Google attempts to remove duplicates from the results, although a few were seen (from different Web sites). |
| Presentation | Very good | Results include thumbnails (20 to a page) along with some associated text, pixel dimensions, file size, file type and source Web site. Once a thumbnail is clicked, a split-screen view provides some simple information along with a link to the full image on a separate page and a warning that the "Image may be scaled down and subject to copyright". Below this information, in a separate frame, is the original source Web page. |
| Support | Good | Specific help is provided for image searching and there are clear notes that images many be subject to copyright. It is, however, difficult to find a way of contacting Google if you would like further help – there is no obvious web form or email address. |
| JISC Digital Media's test words | Very good | Google found the recent BBC image and it brought up a different JISC Digital Media image to the one we chose – but from the same page. Results for our other searches were generally very good. There was a lot of Homer Simpson for the 'Homer' search, but excluding "Simpson" in the advanced search produced some very good images of the classical poet – some of which, like Exalead, could also be detected using Google's new 'face' search. |
Live Search – Microsoft
http://www.live.com/?searchOnly=true&scope=images
In our last search engine review, Microsoft was drawing on Picsearch to provide image searching, although there were hints that it was about to launch its own services. In a short space of time, Microsoft's Live Search has become a significant player, third in terms of size and usage after Google and Yahoo. It delivered some good results in our tests but was a little hampered by its presentation (especially the long scrolling page and need to mouse-over an image to discover more information about it).
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Very big and broad | LiveSearch does not indicate the size of its image index. Judging by its reported number of results, LiveSearch has a very large index, although it reports less than half the number of Yahoo's images and a small percentage of Google's. But note that like these other engines you can only actually view up to 1000 results. |
| Search Options | Average | Live Search's advanced options and user preferences only relate to its general Web search (apart from the adult filter), so you can only conduct a simple search. Once results are found it is only possible to filter images by size, although there are sometimes useful "related" search prompts. Live Search's adult filter defaults to moderate. |
| Performance | Good | Results can be slow on slower connections due to Live Search's scrolling presentation (described below), but they are generally relevant and of good quality. We saw some duplicates, but drawn from different pages or sites, and the occasional adult image. |
| Presentation | Average | Live Search has an unusual way of presenting its results. Instead of pages, it provides all the results (up to 1000) in one long scrolling page. This can cause delays while you wait for the next few rows of thumbnails to download. Information about each image (filename, dimensions, file size and URL) can only be seen when you hover over a thumbnail with your mouse, which can make it more time-consuming to find out about the images. Clicking on a thumbnail opens another screen with the results scrolling down the left and the source Web sites opening up within a large frame. A very small notice down the bottom of the screen warns the user that "Images are scaled down and may be copyrighted". Live Search offers a 'scratch pad' for bookmarking images and a 'feedback' link, which enables users to identify images that are 'inappropriate', 'irrelevant' or 'copyright'. |
| Support | Good | Live Search offers no specific help for its image searching, but it does provide a general feedback form in addition to the individual image feedback described above. |
| JISC Digital Media's test words | Good | Live Search did not find the specific images from the JISC Digital Media and BBC Web sites. It found a reasonably good selection for the other test words, although the 'Homer' search was completely dominated by images from the Simpsons cartoon. |
Lycos - UK (uses Picsearch)
http://www.lycos.co.uk/search/picture.html
Lycos UK is part of the Lycos Europe group of search engines and portals, which is independent from Lycos US. It uses Picsearch to provide its results and, like Picsearch, does not seem to apply limits on the number that can be viewed. The results are slightly different from the main Picsearch engine, which suggests that it may not be quite as up to date. Lycos UK allows filtering by size or colour information. Its results present thumbnails accompanied by pixel dimensions and (unlike Picsearch itself) an indication of the image's Web source. However, when a thumbnail image is clicked the user goes straight to the source Website, without an intervening split-screen. Lycos UK does not provide copyright warnings, but accompanies its image results with advertising and suggests that the user might also want to try a Web search (provided by Ask).
Lycos - US (uses Ask, but changing to Pixsy)
http://search.lycos.com/?tab=multi&cat=images&
Somewhat confusingly, Lycos US is different to Lycos UK (above) and has chosen a different provider of image searching. It currently uses Ask (who also provide its Web results), but there are some indications that it will be moving to Pixsy (below) during 2008. Because Lycos US currently draws on Ask's results, it only displays up to 320 images (20 screens of 16).
Picsearch
Picsearch is a Swedish company and was an early provider of image searching, selling its search services to many of the general search engines. Although it has lost Ask and Microsoft (who've developed their own engines), it still provides services to Lycos UK and Ditto (in this review) along with many other non-English or specialist search engines. Recently it has added general Web searching (provided by Ask) and been growing a video service, which it sells to portals rather than delivers via its main site. Among the image search engines reviewed here, Picsearch is distinctive for two reasons: it seeks to vigorously exclude all adult content, and it does not cap its results, enabling more comprehensive searching. Its performance is generally very good, but it is let down a little by its presentation: requiring a user to click on an image to discover its source.
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Big and broad | Picsearch states that it has over 2 billion images indexed. In our tests it tended to report results close in size to Exalead and Ask, but significantly less than The Big Three (LiveSearch, Yahoo and Google). But note that unlike every other search engine we reviewed, the results do not seem to be capped in any way (we gave up after viewing 25,000!). |
| Search Options | Very good | Picsearch provides both simple and advanced search options, with the latter enabling users to limit searches by/to animations, by colour, or to one of 7 different gradations of size (usefully specified in pixels). There is no adult filter because unlike other image search engines, Picsearch seeks to exclude all adult imagery from its database. Once a search is made there are suggestions for other searches ("Also try") and, if the advanced search was used, it is possible to further refine the search. |
| Performance | Very good | Results are quick and sizeable and of generally good quality. Unlike the other search engines we reviewed, Picsearch does not cap the results, which enables you to do comprehensive searching or filtering. We observed some duplication, but from different pages or sites. |
| Presentation | Average | Picsearch displays 20 thumbnails at a time with only the pixel dimensions and file size displayed. This means that you must click on each image to discover its source and further information (colouration – and number of distinct colours! – with links to the image and its source page). The split screen displays this information on the top, along with a copyright warning, and the original page below it. |
| Support | Good | There is a search help link at the bottom of the basic search and at the top of the advanced search. Picsearch can be contacted and there are emails and forms for getting help or alerting Picsearch to problem images. |
| JISC Digital Media's test words | Good | Picsearch found the JISC Digital Media image when the '.gif' extension was removed. It did not find the BBC image. Results for other searches were generally good, although some were a bit mixed and included duplicates. Picsearch's presentation created difficulties for searching the Picasso painting. It found over 300 results, in a wide variety of different colourations, and because the URLs were not provided, it was necessary to click on each image to find a reliable source. |
Pixsy
Pixsy markets itself as a multimedia search engine, but it draws its results from selected RSS feeds rather than from general web trawling (as the other engines in our list do). As a consequence it offers narrower and more limited content than other engines (i.e. "millions" rather than "billions"). But the flip-side of its selective image gathering is that it is likely to include some content that is out of the reach of other engines. To view and assess Pixsy's sources, search on any term and then click on the 'Source' drop down list. We found Pixsy's performance was generally disappointing. A press release from early 2008 indicates that Pixsy will soon be providing image and video search services to Lycos US (see above).
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Tiny and narrow | Pixsy does not say how many images it has indexed, but our tests suggest the number is quite small. For our search term 'Queen' it found about 400 results, while Google reported nearly 40 million! It looks like Pixsy may only display up to 1000 image results, like most other image search engines, but this could not be tested, since none of our results exceeded 900 images. |
| Search Options | Poor | Pixsy does not provide an advanced search, although you can additionally browse its image and video index by category. By default, Pixsy will search both still images and videos, and its only possible to filter these (by content type, category, and source collection) once an initial search has been conducted. |
| Performance | Poor | Pixsy is quite slow and delivers small sets of results, with images of very mixed quality (i.e. some very good images alongside some very poor images) – this probably reflects its mix of professional and personal image sources. We noted some duplicates and saw quite a few missing images among its results. |
| Presentation | Good | Pixsy displays thumbnails (20 to a page), some very small in size. Accompanying these are a title, the name of its source, and a date (its unclear what these refer to – some are very recent). Clicking on a thumbnail opens a split screen with some information at the top (along with the ability to scroll through related results) and the source Web page displayed below it. The split-screen presentation and a link at the bottom of each results page make reference to copyright and the need to seek permission. |
| Support | Poor | We could find no obvious help information. There is general contact information, but no email for support or feedback and no form. |
| JISC Digital Media's test words | Poor | Pixsy did not find the JISC Digital Media or BBC images and produced mixed or poor results for many of the other searches (nothing for 'Raphus cucullatus'). An exception was 'Eagle'. This produced some good images because its initial results were drawn from a stock photo agency. |
Yahoo
http://images.search.yahoo.com/
Yahoo began its life as a human-made directory rather than a search engine, but during 2003-2004, through some strategic purchases, it established itself as a major search engine player and serious competition to Google. Since our last review, Yahoo has acquired Flickr, a very large photo-sharing site. This has had a significant impact on its image search engine, with Yahoo giving particular prominence to Flickr images among its results. Those analysing Web statistics and trends have noted the significant increase in traffic to Flickr as a result. Few, however, have noted the effects this has had on Yahoo's image search, which are mixed. There are some very high quality images on Flickr, so its inclusion is an enhancement for many kinds of image searches. However, for other searches (e.g. art images or scientific images), Flickr may not be an ideal source and its inclusion risks obscuring more suitable images.
| Criteria | Overall Evaluation | Comments |
|---|---|---|
| Scope | Very big and broad | Yahoo does not indicate the size of its image index. Judging on its reported results, it is probably about twice the size of its nearest competitor (LiveSearch), but small compared with Google's index. But note that you can only actually view up to 1000 results. |
| Search Options | Good | Yahoo provides both simple and advanced searching. The advanced search has Boolean-related options and enables limitation by size, colouration, Web site, or adult filter (which defaults to 'on'). Once a search has been made, it can be further limited by size or colour and there are prompts for other search terms the user might want to "Also try". |
| Performance | Average | Results are very quick and large (although only 1000 can be viewed). In many searches, Yahoo gives prominence to images from the Flickr photosharing site (which it owns). This practice can lead to some mixed results. We observed a significant amount of duplication, especially among the Flickr images. |
| Presentation | Average | The first couple of results pages are usually drawn from a variety of sources, with Flickr images predominating on the following screens and sometimes tailing off in the later results (depending upon the search). Images are displayed 20 thumbnails to a page, along with a filename, dimensions, file size and Web site (for non-Flickr images) or a filename, username and Flickr link (for Flickr images). Once clicked, the split-screen results also differ depending on whether images are from Flickr. Non-Flickr images are tagged with "This image may be subject to copyright" while Flickr images are invariably marked as "© All rights reserved" (even though some we viewed had been assigned Creative Commons licences within Flickr). The split-screen view enables a user to 'View Image' by itself on a separate page or 'Mail to Friend' (if you have a Yahoo identity). |
| Support | Average | Yahoo provides image-related help but does not provide an obvious email address or form for raising questions or providing feedback. |
| JISC Digital Media's test words | Poor | Yahoo did not find the JISC Digital Media or BBC images. Results for some of the other searches were very mixed. Yahoo's promotion of Flickr images works very well for some searches (e.g. Eagle) but not for others ("Chaucer" produced mainly pictures of people's pets). |
6. Meta- searching and 'springboard' searching
Meta-search engines submit a query to one or more other search engine and bring back the results. Such an approach might seem to promise better results, but this is seldom the case – especially when searching for images. Meta-search engines typically offer fewer search options than their source search engines and the results they bring back tend to be slower, smaller in number, and less reliable. Few of the meta-search engines performed particularly well in our tests.
Mamma draws all its results from Picsearch. While it seems to allow you to view all the results, it does not provide the advanced features offered by Picsearch itself. Fazzle picks the top 60 or so results from AltaVista (which uses Yahoo) and Fast (which is now called AllTheWeb and also displays Yahoo results). Ixquick provides less than 100 results. Its sources are unclear. Ithaki, which has featured in previous search engine reviews, no longer seems to deliver image results.
Currently, the following services (all served by InfoSpace) seem only to be providing images from Yahoo. We did not achieve over 100 results from any of them:
However, a fifth InfoSpace service, Webfetch proved something of an exception among the meta-search engines we looked at, delivering thousands of good results (up to 6240 for our test words). Webfetch states that it searches Google, Yahoo, LiveSearch and Ask, although the source of each individual image returned is not named, so the distribution is unclear.
In addition to meta- search engines are the services we're termed "springboard" search services. Although they're often classed as meta-search engines, they are more like 'springboards', since they only pass on the user's search to one search engine at a time. Since they only offer very simple search options and are doing little more than passing a message on, they might seem like a waste of time.
Photoseek gives its users the option of searching Photos.com, ClipArt, Ditto, Excite, Lycos or AltaVista. It simply passes on the request to the selected search engine and then frames the results that come back.
Search 22 Picture and Image Search Engines offers a much better spring-board to image searching than Photoseek or any of the proper meta- search engines listed above. Its twenty-two image search engines are a mixed bag, but many of the main engines are included. While Search 22 does not offer the sophistication available when these other engines are searched directly, it does enable the user to make a very quick check across a number of different services – which could be useful in ascertaining which engines are worth searching in a more in-depth fashion.
7. Content-based image retrieval
The image search engines reviewed above rely primarily on text and context strategies. There have been a number of attempts to build content-based search engines. Content-Based Image Retrieval (CBIR) considers the characteristics of the image itself, for example its shapes and colours. To date, these attempts have been experimental and have often been limited to individual collections. Well-publicised efforts have included Columbia University's WebSEEk project and IBM's QBIC (Query by Image Content), which can be seen in action on the Hermitage Museum website. For a UK example, see the IMAGINE collection.
It is likely that some search engines have been employing a form of content analysis to identify potential adult images (e.g. percentages of flesh-coloured tones). However, with Exalead and Google's recent face searching we are seeing CBIR make a more explicit entry into mainstream searching - albeit in a fairly small and controlled way. Exalead and Google are almost certainly combining both CBIR and text analysis techniques: identifying common features of faces, but relying on textual clues to determine which individuals these faces belong to.
Wikipedia's entry on CBIR maintains a good, current set of links to further CBIR examples, software, and research.
8. Conclusion
Image search engines attempt to give access to a wide range of images available on the Internet. For those used to viewing well-indexed collections of high-quality images, the results of the large automated image search engines will probably disappoint. The mixed quality of their offerings is not surprising, since this reflects the randomness and unevenness of the Web. The frequent irrelevancy of their results is also explicable, since the automated engines are having to guess at their images' visual content using indirect textual clues.
Anything then, that enables the user to have more control over their image searching is helpful. The ability to filter a search – to include and exclude terms based on search terms or image characteristics – will be a help. Many users will want to exclude adult imagery from their search results, but it can also be very useful to limit by file type, file size or colour. An image search engine also needs to return a reasonable number of results, since in any given search a fair proportion of the images found are likely to be irrelevant or of insufficient quality.
Here the meta- image search engines are likely to fail their users. They deny them some of the sophisticated filtering available from other engines and will typically return a small handful of results. While a 'springboard' service like Search22 can be useful in identifying the best search engine to try for a particular task, the final searching will be better done directly, using the main engine itself. Of these, Google, Exalead and Picsearch seemed to perform the best in our tests, but for different reasons. Some of the others (particularly Ask, Live Search and Yahoo) will also be worth trying, although the user should bear in mind some of the limitations we've identified in this review.
The Internet is a very fast-changing environment, so it is likely that some of the information in this review will quickly date. As we suggest above, the reader
