Search system, search method, and program

A search method comprising:

• storing a feature quantity of a first image to be searched, which has been input in a learner, in a database, the learner calculating a feature quantity of the first image that is input and outputting a score of the first image based on the feature quantity, the score indicating probability that is output by the learner and the box information indicating a position if a bounding box in the first image;
• inputting a second image in the learner;
• and
• searching for the first image to be searched that is similar to the second image in the feature quantity based on the database;
• wherein
• the learner includes an intermediate layer, a first output layer, and a second output layer,
• the intermediate layer is disposed before the learner branches into the first output layer and the second output layer and calculates the feature quantity,
• the first output layer outputs the score based on the feature quantity calculated by the intermediate layer before branching,
• the second output layer outputs the box information based on the feature quantity calculated by the intermediate layer before branching;
• searching for the first image that is similar to the second images in the score, and, among a plurality of retrieved first images, searching for the first image that is similar to the second image in the feature quantity;
• and
• searching for the first image that is similar to the second image in the box information among the first image that is similar to the second image in the score, and, among the plurality of retrieved first images, searching for the first image that is similar to the second image in the feature quantity.

A non-transitory computer-readable information storage medium for storing a program that causes a computer to:

• store a feature quantity of first image to be searched, which has been input in a learner, in a database, the learner calculating a feature quantity of the first image that is input and outputting a score of the first and a box information of the first image and a box information of the first image based on the feature quantity, the score indicating probability that is output by the learner and the box information indicating a position of a bounding box in the first image;
• input a second image in the learner;
• and
• search for the first image to be searched that is similar to the second image in the feature quantity based on the database;
• wherein
• the learner includes an intermediate layer, a first output layer, and a second output layer,
• the intermediate layer is disposed before the learner branches into the first output layer and the second output layer and calculates the feature quantity,
• the first output layer outputs the score based on the feature quantity calculated by the intermediate layer before branching,
• the second output layer outputs the box information based on the feature quantity calculated by the intermediate layer before branching;
• search for the first image that is similar to the second image in the score, and, among a plurality of retrieved first images, search for the first image that is similar to the second image in the feature quantity;
• and
• search for the first image that is similar to the second image in the box information among the first image that is similar to the second image in the score, and, among the plurality of retrieved first images, search for the first image that is similar to the second image in the feature quantity.