Global color contrast detection and saliency map segmentation method

Abstract

The present invention belongs to the color image processing technical field and relates to a global color contrast detection and saliency map segmentation method. The method of the invention includes saliency detection and saliency map segmentation; according to the saliency detection, a global color contrast method is adopted; according to the saliency map segmentation, a dynamic threshold method is adopted. According to the method of the present invention, a preliminary saliency map is obtained through extracting the features of global color contrast; the saliency of the preliminary saliency map is improved through using color dispersion and spatial information; and with the saliency map generated in the above steps adopted as an initial value, a threshold value which is dynamically updated by iteration is adopted to segment a plurality of salient targets. As indicated by the experiment result of a published Achanta database, the performance of the method of the invention on an ROC (receiver operating characteristic) curve is superior to that of methods in literatures and reports, and the method of the invention is applicable to multi-target detection and segmentation problems.

Description

technical field [0001] The invention belongs to the technical field of color image processing, and in particular relates to a salient detection and segmentation method of global color contrast. Background technique [0002] At any one time, there are a large number of visual stimuli in the environment. After a long evolution, the Human Visual System (HVS) can quickly extract objects and regions of interest from complex environments to reduce the complexity of visual signal processing. These local areas that attract visual attention are usually called saliency regions. The image obtained after normalizing the saliency of each area in the scene is called a saliency map. [1] . [0003] In image processing and computer vision, a robust and accurate salient region automatic detection algorithm has important practical value. This is because from the perspective of computational complexity, some image processing tasks often cannot process all the visual information in the visua...

AI Technical Summary

Problems solved by technology

The main disadvantages of these existing reported algorithms are: algorithms based on local contrast class usually calculate the saliency by detecting the edge of the image, so only the edge of the target can produce high saliency; while some other algorithms can only detect Global maximum saliency, so it is not suitable for multi-target detection problems; algorithms that ignore the spatial relationship of parts of the image do not effectively use spatial information, etc.

Images