Significance detection method based on discrete cosine coefficient multi-scale wavelet transform

Abstract

The invention discloses an image vision significance calculation method based on discrete cosine coefficient multi-scale wavelet transform, belonging to the technical field of image processing. The method includes scaling an input image size; three color channels and intensity channels of generalized red, green and blue are calculated. Calculating weight coefficients of red, green, blue and intensity channels; the amplitude matrix and sign matrix of DCT for red, green, blue and intensity channels are calculated. Multi-scale wavelet transform is used to compute the multi-scale amplitude matrices of red, green, blue and intensity channels. Multi-scale channel salience maps for red, green, blue and intensity channels were calculated. The multi-scale channel salience map is synthesized into the multi-scale spatial domain visual salience map. Fusion saliency map is generated by selecting better saliency map according to saliency evaluation function. The final saliency map is generated by optimizing the central bias of the fusion saliency map. The invention can quickly and effectively calculate the saliency value of the image, and the salient objects in the obtained saliency map are complete, and the background interference is small.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a saliency detection method based on discrete cosine coefficient multi-scale wavelet transform. Background technique [0002] The human visual system has a bottom-up visual attention mechanism, which can quickly grasp and extract objects of interest, greatly reducing the occupation of brain nerve resources. The selective visual attention mechanism processes information in a serial manner of neurons, allowing only a small amount of perceptual information to enter the visual higher cortex, thereby highlighting salient objects and ignoring surrounding background areas. Visual attention mechanism According to the attention mechanism, it is divided into scene-dependent or bottom-up visual attention and task-dependent or top-down visual attention. The visual saliency detection algorithm simulates this visual attention mechanism by computer, highlights the salient target, supp...

AI Technical Summary

Problems solved by technology

[0005] Recently, a class of saliency detection methods based on deep learning has emerged. These methods include "Deep visual attention prediction" proposed by Wang et al. in 2017 and "Efficientsaliency detection using convolutional neural networks with feature selection" proposed by Cao et al. in 2018. The saliency map obtained by the class method has high accuracy, but the training process is complicated and the calculation speed is slow

Images