What is Image Search?
Image search is a function found in some search engines that lets you search for images on the Internet based on some search criteria, such as a description of the image or related keywords, as well as specifying size, color, aspect ratio, among other filters; it can also be a function of local applications that search for images on your computer.
Where You Can Search for Images on the Web
This function is found in several general-purpose search engine websites, such as Google, Bing, Yandex, Kagi, and Brave Searcb, which offer web search that let you search for webpages, and have a different mode found in a tab that lets you search for images on the web, or, more specifically, for the webpages that have said images.
There are also some specialized web search engines, like GifCities, that only search for gifs found in archived GeoCities websites.
Besides these, we also have several websites that specialize as image-sharing platforms. There's Instagram and Pinterest, which are the major image-based social media. There are various platforms for artists and photographers, each with their own niche, such as Flickr and 500px (photography), ArtStation, DeviantArt, FurAffinity (furry art), Pixiv (anime art), Dribbble (design), and Ravelry (knitting and crochet). There are also websites specialized in royalty free images or copyleft images that you can use for your own projects, such as Wikimedia Commons, Pixabay, Openverse, and OpenGameArt. A feature found in some of these that isn't found on the web in general is the use of tags and hashtags to categorize images.
How Image Search Works
In general, image search engines can't search for an image based on what is contained in the image. For example, if you search for "cat," the search engine can't actually tell whether there's a cat in the images it shows you, so describing the image doesn't really work.
Instead, what image search engines do is try to match your query with words in the same webpage as the image. If you type "cat," any webpage that contains the word "cat" and also contains an image will be eligible to show in the results, even if that's not an image of an actual cat. Just the fact the word "cat" is there somewhere will be enough. Because the image search engine isn't searching images by the description of the image, it's searching for webpages by text, then showing the images in those webpages.
Another example: if you want to find a picture of a "white cat," and there is an image of a white cat in a webpage, but the webpage doesn't contain the words "white cat," then the image search engine won't be able to find it for you. If you search for "white cat on a table wearing a hat," which sounds like a prompt someone typed in an AI image generator, the image search engine won't be able to find any results because nobody will have describe their own images with such precision, even if they took an image like that.
On image-sharing websites, like Flickr, users are more likely to at least describe the main element of a photo, but they still wouldn't describe their pose and accessories like that.
Instead of describing what you want to find, you need to type exactly the same words that you think will be in the webpage where the image will be found.
Nowadays, we have AI image generators, which means extremely descriptive prompts. There are websites where AI-generated images are posted, and these prompts will be on the webpage as well. As a consequence, if you type something very specific, many of the results will actually be AI-generated images, because they're from the only webpages that would have such specific descriptions for images that they would match very specific queries. Another case are stock images. Websites that sell stock images also have an interest in describing their products extremely specifically so that the images can be found by search engines. Lastly, we have products. Online shopping websites also describe their products in an extremely specific manner.
As you can imagine, it's very hard to find a photo that contains something you describe that doesn't come from someone trying to sell you something, because the only people who would describe a photo in such detail are people in the business of describing images.
Filters
Many online image search engines let you specify some extra filters besides the text query. Let's take a look at what's available before understanding how it works.
On Google image search, you can select a size (large, medium, or icon), a color, a type (clip art, line drawing, or GIF), how recently it was posted, and it's licensed under creative commons.
On Bing image search, you can select a size (small, medium, large, extra large), or specify that the image should be "at least" an amount of pixels. You can specify a color, a type (including photograph and transparent this time), a layout (square, wide, tall), you can search for images of people's faces or from shoulder up specifically, specify the date of the image, and choose one of various creative common licenses, or public domain images.
On Yandex image search, you can specify a size vaguely (large, medium, small), or specify a size in pixels exactly, you can specify an orientation (landscape, portrait, square), a type (photo, with white background, drawings, people, demotivators (what?)), specify a color, a format (JPEG, PNG, GIF), and choose whether you want recent images, or wallpaper images, or neither.
On Brave Search, there are no filters for images, just settings such as safe search.
Searching Images by Specifying Exact Size
As noted above, of Google, Bing, and Yandex, only Yandex image search lets you specify the exact size of an image. Google doesn't let you specify anything. Bing lets you specify the least amount of pixels, which means if you type 1600x900, you could get an image of size 1600x3200, with a completely different aspect ratio from what you want.
It's worth noting that even though Google lets you specify "icon" as an image size, Google has some enormous ideas for what counts as an icon. Google simply put doesn't index images that are extremely small (e.g. 16x16px), likely because most of the time these are actual icons used around a webpage and would be difficult to describe with other words available on the webpage, and people don't really search for these things to begin with.
Search Images by Color
Most image search engines let you specify the "color" of the image. What this means is that the image search engine's algorithms detected one color or another as the primary color of the image, or a significant color of the image, and stored that in the search engine's index.
If you search for "cat" and select blue, for example, you won't get pictures of blue cats, because blue cats don't exist, you'll get pictures of normal cats with a blue background.
Search for Transparent Images
Some image search engines have the special option "transparent" in the color filter dropdown. Selecting transparent limits results to PNG, WEBM, and GIF image formats that support transparency (JPG doesn't support it).
Note that sometimes you may find a fully opaque image despite using this filter because the image was saved in a way that lets it store transparency data (the alpha channel), but doesn't actually have any transparent data in it.
Search for Black and White Images
Some image search engines have a special option "black and white" in the color filter dropdown. Selecting it limits results to black and white (monochrome) images.
Search for Aspect Ratio
Some image search engines let you specify the aspect ratio (also called the "orientation" or "layout") of the image. As we've seen, sometimes it's landscape or portrait, other times it's wide or tall. The meaning is the same. And square is square.
Search by Type of Image
Some image search engines let you specify if you want to search for photos, clip art, line drawings, etc.
Yandex lets you search for demotivator images, which is an old meme format in which the background of the image is black, and then there's an image in it with a caption in white text under it.
The images aren't self-described as photo or clip art, which means the image search engines need to detect that somehow with an algorithm. The demotivator filter, for example, will also match images that just have a black background with white text on it.
Search by Date of Image
Search engines may let you filter the results by how recent they are, but this filter is rather unreliable. For example, if you specify "last week," does that mean the search engine found the result last week, or that the image was posted last week? Naturally, all images are posted before the search engines can find them, and they could take months to find an image. Additionally, the only way we have to tell when an image was posted is by the date contained in the webpage, and the owner of the webpage could write anything there. They could say the image was posted in 300 B.C.E., or even posted in the future, in 2999.
On False Negatives
As you may have noticed from the sections above, no matter how good an image search engines is, it's not perfect, and it will mis-categorize images sometimes. Even if you knew there is a mainly blue photo of a cat posted last week, that doesn't mean te search engine will be able to find it with those specific filters, because it may have categorized the image as a line drawing by mistake, or associated the color white with the image instead of blue.
These scenarios are called false negatives, where we deem something doesn't pass a criterion, even though it does.
These filters are pretty good for removing unwanted images if you're trying to find a random image that fits a criteria, but if you're trying to find a single specific image, using these filters may end up excluding the result you want.
Leave a Reply