Collaborative visual saliency detection method based on double-encoder generative adversarial network

Abstract

The invention provides a collaborative visual saliency detection method based on a double-encoder generative adversarial network. The method comprises the steps of: constructing and pre-training a double-encoder generative adversarial network model, wherein the pre-trained parameters are used for a generative adversarial network model; inputting the collaborative saliency data into a classification network module as a group of images, extracting multi-scale group-level image semantic category features, and enabling a multi-scale semantic fusion module to fuse the multi-scale group-level image semantic category features into inter-group saliency features; sequentially inputting the grouped input images into a saliency encoder one by one to obtain a single saliency feature; performing pixel-level addition on the single saliency feature and the inter-group saliency feature to obtain a collaborative saliency feature, and inputting the collaborative saliency feature into a decoder for decoding to obtain a detection image; and detecting the trained generative adversarial network model by utilizing the collaborative saliency data set. The method is small in model parameter, simple in training and detection operation and high in detection precision, and the efficiency is improved.

Description

Technical field[0001]The present invention relates to the technical field of synergistic significance detection, and more particularly to a synergistic visual significance detection method based on a dual encoder generating counterfeit network.Background technique[0002]With the continuous development of Internet and multimedia, a large number of images and video data are accompanied by our daily lives, how to use existing multimedia technologies to quickly and efficiently obtain useful information. Nowadays, a collaborative significance test technology is a computer visual technology that mimics human visual attention mechanism. It can find a similar significant goal from a set of images, and find each image in an associated image. Public significant goals. Such methods can effectively obtain the information you want to simultaneously filter redundancy information in the image, thereby achieving the efficiency of reducing computer storage and improving calculations.[0003]There are t...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the technical problem that the traditional co-saliency detection method has insufficient sample labels and the saliency features between groups are not good enough to mine the semantic consistency of images between groups, this invention proposes a dual-encoder-based generative adversarial network The collaborative visual saliency detection method, which jointly trains salient features between groups and single salient features, has strong versatility, has few model parameters, and has high detection accuracy. The problem of insufficient label samples

Images