Image Search is a feature that allows you to find out where your image has been used on the Internet. Also, many people look for information on pictures to find out what is depicted on them. Searching for an existing image is also called a reverse search. It is available in all modern search engines. If you want to find a person by a picture, and he is not a public figure, with a high probability, the search engine will only return a mass of similar pictures, and not exact matches with links to social networks. Regardless of the type of photo and its quality, you will always find the information you need. Direct links to similar pictures, exact matches and addresses of thematic pages are displayed in the search results.
In our article, we will consider ways to reverse image search in the most popular search engines. We took three search engines and our given image and started searching. The most famous search engine Runet Yandex has a function of recognizing objects in a photo, which works perfectly both from a phone and from a computer. Yandex is recommended if you want to find any Russian-language content. Searching people’s photos on Google doesn’t work very well. Google will show the name of the object in the photo, all available photo extensions and the most suitable websites for the given topic. Bing Visual Search lets you search using images instead of text. You can use visual search to find similar images, products, pages containing images, and even recipes. So, let’s take the image as an example and search for the image in well-known search engines.
In the first example, Yandex found numerous photos of the wanted person from different sources (only two of the top results featured outsiders), and the results differ from the original image, but show the same person. Google found nothing at all, and Bing showed only one result with the same person (fifth image, second row).
Look at the results of “Yandex”, which 100% immediately gives the right person, in second place – Bing, where the search has several unique functions, such as selecting a specific area of the image for search.
Google is good for the simplest reverse lookup. For example, identifying famous people in photographs, finding the source of images, determining the author of a work of art, etc. However, if you want to find similar images (not exact copies), you will be disappointed.
For testing different methods and mechanisms for finding reverse images. Let’s take a few images that constitute different types of research, including both original photos (not previously uploaded to the Internet) and reworked ones. The fact is that when an image is published on the Internet, search engines index these photos and integrate them into their results.
On the building in Nizhny Novgorod, the best results were shown by the image search of “Yandex”. In this case, Yandex recognized this house without any problems. He found photos taken from the same angle, and also found from other angles. Yandex also easily recognized the white SUV in the foreground of the photo as a Nissan Juke. After all, in the most difficult isolated search for this image, Yandex was unable to identify a nondescript gray trailer in front of the building. Some of the results look like the source image, but none of them are true.
Compared to these results, the output of Google and Bing looks simply ridiculous, although Google correctly classified the trailer as a travel trailer. If the image search does not give results, in some cases simple tricks help:
mirror image of a photo;
application of color filters;
removing unnecessary elements from the frame that can complicate the search.
In addition to the standard image search, there are a number of tools available to assist with online investigations. There are specialized tools for processing certain types of photos. For example, Cornell Lab’s Merlin Bird ID program is extremely accurate at identifying the type of bird in a photo or suggesting possible options.
Or FlagID, where you can manually enter information about the flag and find out its origin. If there are symbols of an unknown language in the photo, you can manually repeat them using the Google Translate handwriting tool.
As detailed in this Twitter thread, you can pixelate or blur elements of a photo to trick the search engine into focusing only on the background. In this photo of press secretary Rudy Giuliani, the exact image makes it difficult to tell where the picture was taken.
But if you blur/pixelate the woman inside the image, Yandex is able to analyze other elements of the image: chairs, paintings, chandeliers, carpets, wall patterns, and so on.
After that, Yandex knows exactly where the image was taken: it is a popular hotel in Vienna.
Reverse image search engines have come a long way in the past decade, and progress continues at a rapid pace. Progress is facilitated by the large growth of the search base. Large Internet companies have managed to convince users to place personal photo archives on their hosting, on which artificial intelligence is trained. It is for this purpose that Google Photos and Yandex.Disk offer unlimited photo storage for free. This is a lot of material for machine learning. They predict that Facebook or Instagram will soon have a publicly available facial recognition program, which will deal a serious blow to privacy on the Internet, but will also increase the effectiveness of digital investigations.