Research On Visual Saliency Identification Technology

With the continuous development of science and technology, visual has become a major carrier of information in our life. The number of visual increases explosively. However,the computer processing speed for visual cannot catch up with its increasing speed. We dream that machine can obtain useful information from these visual data quickly and accurately like human beings. Therefore the research on visual saliency algorithms are carried out to achieve the goal.This thesis focuses on a visual saliency algorithms based on multiscale and super-pixel segmentation.Firstly, analyzing and comparing visual attention mechanism, visual feature and some typical visual saliency detecting algorithms. So a visual saliency detecting strategy is established which based on visual processing mechanism in the Bottom-Up way, taking into account the characteristics of the visual color and spatial information, and combined the traditional Region Contrast(RC) algorithm. Secondly, an improved algorithm based on the Simple Linear Iterative Clustering(SLIC) superpixel segmentation is produced after analyzing and comparing some typical superpixel segmentation algorithms. The improved algorithm can retain the good result of the original SLIC segmentation algorithm, only processed in under-segmentation region. The simulation results show that compared with the original SLIC algorithm, the improved SLIC algorithm works better. Afterwards, a visual saliency algorithm based on multiscale and super-pixel segmentation is studied. The multiscale space of the original visual is obtained by building Gaussian pyramid. The superpixel sets is done for visual in in every subscale using SLIC. Then, the color and space distance of the super-pixel sets are calculated in each scale. And the salient region is segmented. The simulation results show that the proposed algorithm is effective. Finally, the original SLIC algorithm in the model is replaced by the improved algorithm, which again proves that the improvement of SLIC algorithm is effective.