Visual saliency detection method by fusing dual-channel color contrasts

Abstract

The invention relates to a visual saliency detection method by fusing dual-channel color contrasts. The method comprises steps of: firstly extracting Gaussian scale images of an input image, dividing each layer of Gaussian scale image into a plurality of superpixel regions which are not mutually intersected and with an approximately equal area; and respectively extracting three color component images of each layer of Gaussian scale image in a CIELAB space by using a color transform method, and according to the three color component images, distributing a color saliency value based on the CIELAB space for each pixel; then respectively extracting the three color component images of each layer of Gaussian scale image in a RGB space by using the color transform method, and according to the three color component images, distributing a color saliency value based on the RGB space for each pixel; and at last fusing the color saliency values of the two spaces, so as to obtain the final saliency value of each pixel in an input image. According to the visual saliency detection method by fusing the dual-channel color contrasts, the limit of the single-color-channel detection can be effectively overcome, and the robustness of the saliency detection can be improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and specifically relates to a visual salience detection method that combines CIELAB color channel contrast characteristics and RGB color channel contrast characteristics. Background technique [0002] The visual selective attention mechanism can break through the bottleneck of information processing, making it easy for humans to judge locally salient regions. In computer vision, saliency provides a biologically-inspired processing framework for artificial vision systems, enabling the prioritization of computational resources required for image processing and analysis. At present, visual saliency has aroused widespread interest in the fields of computer vision and image processing, including image segmentation, adaptive compression, image content editing, etc. [0003] Currently, visual saliency detection models can be roughly divided into two categories. One class is based on biologically inspire...

AI Technical Summary

Problems solved by technology

However, the current computational models with excellent performance usually perform saliency analysis in a single scale space, so it is difficult to deal with saliency detection with complex background textures or complex target textures.

Current computing models based on multi-scale spaces often face limitations such as blurring of prominent target outlines, so the recall rate and precision are unsatisfactory.

Images