Extraction of Salient Region Based on Visual Perception

摘要

In order to better analysis and understand digital images, a salient region extraction algorithm based on visual perception is proposed. First, a multi-scale difference of gaussian filter is used to the image, which simulates the center-peripheral response of the human visual nerve cell; Then, we use the pulse cosine transform to extract the edge information of the image, simulate the side inhibition process of the nerve cell, and obtain the image feature maps at different scales. Finally, the threshold segmentation and regional expansion of the feature graph are used to construct the focal window in the region with the most prominent image features. The experiment was performed on 500 images of the salient object database provided by Microsoft Research Asia (MSRA) using the proposed method, take β~2 = 0.3, the value of F-measure is as high as 0.816. Results show that the method can be effectively to extract salient area of the image with different content, different target location and size, and has the location and size on the adaptability.