Image And Video Saliency Detection Based On Visual System

Saliency detection is mean to detect the salient regions in images or video sequences, namely detect the regions which are most interested to human’s eyes, and these regions always contain mainly information of the whole image. By detecting the salient regions of the image, we can be more rational allocation of computing resources, so we can streamline the computation and improve the accuracy in image analysis, understanding and processing. Image saliency detection is widely used in computer vision.We pay mainly attention to image saliency detection algorithms and their application, and also the saliency detection works in video analysis and processing. The main work and innovation is as follows:1. Improve an image saliency detection algorithm based on visual system, and apply it in representative image selection to get better results.The improved algorithm uses a new way to change the color space and calculate the mean value of the image depends on its statistical characteristics, so that we can get a better saliency map. Then calculate the minimum bounding rectangle of the salient region, and interception a new processed image from the original input image corresponding to the minimum bounding rectangle. After all, apply the the new processed image in the representative image selection processing and select more comprehensive representative images of a network album in less operation time.2. Design a target-driven saliency detection algorithm, and the experimental results indicate that this algorithm can quickly detect the subjective salient regions.The target-driven saliency detection algorithm consults human’s subjectively paying attention processing. Firstly, the user should choose a target image that should be paid attention to. Secondly, calculate the difference between the pixels of the test image and the average value of the target image in L*a*b* color space. We define that the smaller the difference between a pixel of the test image and the average value of the target image is, the higher the saliency of this pixel.3. Improve a video saliency detection algorithm based on visual system, and the experimental results indicate that the improved algorithm can detect the salient regions in video sequences more exactly. The improved algorithm combine static saliency map and the dynamic saliency map of a video sequence to obtain the final saliency map.