Instance retrieval method based on depth activation salient region

Abstract

The invention discloses an instance retrieval method based on a deep activation salient region, and relates to visual instance retrieval. The method comprises the following steps: 1) performing modeldesign: a model comprises a forward propagation module, a mode positioning module and a feature extraction module; 2) for a given image database, taking each picture in the database as the input of the model, and extracting an output instance positioning result and a corresponding instance-level feature; 3) for each query picture, taking the query picture as input of a forward propagation part ofa depth mode mining module, extracting regional instance features, and performing similarity comparison on the instance features and all instance-level features extracted from the database picture bythe model to obtain a similarity comparison result; wherein the area with the highest similarity in each database picture is the result of instance retrieval on the picture, the similarity corresponding to the area is the similarity of each picture, all pictures in the database are sequentially arranged from high to low according to the similarity, and the result of instance retrieval of the wholedatabase is obtained. The invention can be applied to intelligent retrieval and video editing of video media.

Description

technical field [0001] The invention relates to visual instance retrieval, in particular to an instance retrieval method based on depth-activated salient regions, which can be applied to Internet companies, intelligent retrieval of video media, video editing and other fields. Background technique [0002] Visual instance retrieval (hereafter referred to as "instance retrieval") is often viewed as a subtask of image retrieval. The retrieved object is a visual instance, usually given in the query image in the form of a box annotation. For example, to retrieve a specific person or a specific car. Instance retrieval needs to find out the picture containing the query instance in the image database, and give the location of the instance in the picture. Instance retrieval can be widely used in intelligent product retrieval, video editing and other fields. It is a common and basic technology in the current image data processing field to support tasks related to the instance by ret...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to use depth-activated salient regions to solve the problem of locating instances of unknown categories, improve the robustness of instance object types in instance retrieval tasks, and provide solutions that can be applied to intelligent product retrieval, video editing, and other fields. An Instance Retrieval Method Based on Depth Activation Salient Regions

Images