Visual significance detection method based on self-learning characteristics and matrix low-rank recovery

Abstract

The invention proposes a visual significance detection method based on self-learning characteristics and matrix low-rank recovery, and the method comprises the steps: adaptively learning a group of characteristic extraction templates according to the raw data of an input image, carrying out the convolution of the inputted image through the group of characteristic extraction templates, and obtaining a characteristic matrix of the inputted image; carrying out the low-rank recovery of the characteristic matrix, and obtaining a low-rank matrix and a sparse matrix through decomposition, wherein the sparse matrix represents a salient region of the inputted image; obtaining a significance value through solving the 1-norm of each column of the sparse matrix, and obtaining a visual significance detection result of the inputted image through Gaussian blur processing. The method is small in calculation burden, is high in detection efficiency, remarkably improves the accuracy of visual significance detection, and can carry out the visual significance detection of various types of images. The visual significance detection result has important significance to the image classification, image compression, and target recognition.

Description

technical field [0001] The invention relates to the technical field of visual saliency detection, in particular to a visual saliency detection method based on self-learning features and matrix low-rank restoration. Background technique [0002] The essence of visual saliency detection is to calculate the degree to which each part of an image attracts people's visual attention. In recent years, with the advent of the era of big data, people's demand for data such as images has been increasing, and it is very necessary to obtain more effective information from them quickly and preparedly. Visual saliency detection can quickly locate the more attractive areas in the input image, which can significantly reduce the data of massive input images, and selectively process each scene area in different orders and strengths, thereby avoiding computational waste and reducing the cost. analysis difficulty. [0003] In the process of visual saliency detection, feature extraction is an im...

AI Technical Summary

Problems solved by technology

The first type of feature extraction method usually uses multiple feature extraction operators while ensuring the feature integrity of the detected input image, but there is a large amount of information redundancy between these feature operators, resulting in a waste of computing resources.

Although the second type of feature extraction method is not manually set but learned from training samples, due to the limitation of the range of training samples, the learned feature extraction template cannot be effective for any image, and there is a problem of adaptability

Images