Three-dimensional image visual saliency detection method based on local comparison and global guidance

Abstract

The invention discloses a stereoscopic image visual saliency detection method based on local comparison and global guidance. The method comprises the following steps of selecting a left viewpoint image, a depth image and a corresponding subjective visual saliency map of the original three-dimensional image to form a training set; constructing a neural network, inputting the training set into a neural network model for training, and calculating a loss function value between each saliency prediction graph and a corresponding subjective visual saliency graph in the training set; and utilizing thetrained convolutional neural network to perform prediction processing on a to-be-detected three-dimensional image to obtain a visual saliency map, thereby realizing saliency detection of the image. Features of an RGB image and a depth image are extracted by using the convolutional neural network. A global up-sampling block is used for extracting global information, and the up-sampling block is used for learning local comparison features and gradually fusing comparison information. Meanwhile, the latest progress in the aspect of deep learning, such as an attention model, is utilized, so that the network pays more attention to a remarkable target, and a deeper network architecture is allowed to realize more accurate detection.

Description

technical field [0001] The invention relates to a visual salience detection method, in particular to a stereoscopic image visual salience detection method with local comparison and global guidance. Background technique [0002] Visual saliency is an important feature of the human visual system in processing visual information. It is the cognitive process of selecting relevant regions while obtaining the most important visual information from a visual scene. As an important and challenging problem in computer vision, saliency detection has attracted a large number of researchers in the past few decades because it can be used in various multimedia processing applications such as object recognition, image retargeting, Image compression, object tracking, defect detection, abnormal event detection and identification and other tasks. Saliency detection methods are generally divided into human gaze prediction methods and salient object detection methods. The first goal is to det...

AI Technical Summary

Problems solved by technology

This may ignore the intrinsic relationship between color features and deep features

Images