18 MAY 2018

Image search: what is it, how it works, the advantages

An image search is a service that can filter images based on various parameters such as size, type of image and others. An image search engine can work in two ways:

  • you can search for a word and obtain all the images related to that word;
  • you can upload an input image (through a specific search box) and you obtain the images which are visually most similar to the original one.

Search engine for finding images, how it works

Generally, a basic image search engine gives some “labels” to each image. If you google one or more words, the search engine will show all the images that have, among their labels, that label or one close. In addition, if the label is semantically close or is the same of the searched word, the associated image will be shown first in the result list.

For example, if I google the word “cat” in the search box, the first results will be the images with the “cat” label, then the ones labeled “felines” and so on.

When we upload an image, first of all, that image will obtain some labels and then what we have already said will be repeated for each label of the image, so the first results will be the images with more labels in common with the one we uploaded.

So, it is easy to understand that a search to find images works better when it is likely that the image is in many locations on the web. For example, images of popular places or VIPS generate more results than personal images such as the photo of a friend or relative.

Image search: Consulthink case study based on Amazon Rekognition

Amazon Rekognition is one of the cloud computing services among those of Amazon Web Services (AWS). Specifically, Amazon Rekognition is a service that can identify, through sophisticated machine learning algorithms, objects, people, places and activities from images and/or videos in a short time.

Consulthink exploited the great potential of Amazon Rekognition to create an image search engine for an important customer. The customer’s need was to have a storage of images and to easily access them through specific research.

Thanks to Amazon Rekognition, Consulthink gave “labels” to the images that were uploaded into storage. Consulthink designed software that allowed the uploading of one or more pictures in storage and also allowed the user to search for archived images. The search was textual (it was necessary to enter one or more words) and provided a list with image previews. The images found could be downloaded onto a PC in the original format. Also, the autocomplete function facilitated the search: while the user was writing a word, the system showed all the labels in the dataset containing the letters just inserted and the first 5 most recurring labels in the dataset. The user could click directly on those labels to do the search.

Search engines for finding images: Facebook, Gazopa and Google

The best-known image search engine is absolutely Google Images. There are, however, others: Facebook, GazoPa, Yahoo Image Search, Instagram, Pinterest and Corbisimages. These image search engines make it possible to do searches by setting multiple filters (such as size, color, etc.), options that Google does not offer. Besides, these image search engines look at websites where Google is not able to.

As you can see, social networks, like Facebook, work on image search engines. The basic principle of image search engines of social networks is that if a person is tagged several times in the various photos uploaded, this will be recognized later even if he/she is not tagged again.

Elaborated by Lucia D’Adamo, in collaboration with Serena Banci, supervised by Marco Pirrone