Region-based Research On Image Segmentation And Visual Saliency Detection Algorithm

With the rise of the Internet technology and the popularity of smart phones, digital image has gradually become the indispensable information carrier in people’s daily life. In this thesis, the main topics are the segmentation of digital and the visual saliency detection algorithm based on regions. is used to segment an image by its different attributes. Regions of the same attribute will be segment into an region for better understanding of images for the computers. Detection of visual saliency detects the salient regions of the image according to the visual attention mechanism, which is used in the self-adaption process of the image.According to these two basic issues above, image segmentation with bag-of-pixels and saliency detection based on moving visual focus are proposed in this thesis.●Image segmentation with bag-of-pixels subsamples the images by taking multiple pixels as a pixel bag. And then the subsampled image is segmented by the GB segmentation method. At last the original image is segmented according to the subsampled image. Comparison of the performance between our algorithm and GB algorithm is conducted in this thesis. And the result indicates that the bag-of-pixels method is better than GB algorithm.●By establishing a saliency gravitation model, saliency detection algorithm based on moving visual focus acquires the visual focus of the image and the saliency maps by weighted calculation of the visual focus. In the gravitation model, the geometric center is taken as the initial visual center of the image. And during the iteration, distance and direction moved is calculated according the gravitation model to acquire final visual focus. By working on the largest existing salient image set, this method is compared with other methods. It turns out that the detection method in this thesis is better than the10well-known methods.