Salient Region Detection And Application Based On Visual Perception

Reliable high-precision image saliency detection technology can greatly reduce the redundant information in the image,and improve the accuracy and efficiency of image processing.Although there are many theories and methods about saliency detection,the salient map computed by existing algorithms are more accurate,but still can not achieve the perfect result.In order to improve the detection speed and the accuracy of saliency detection result.We have done the following work focuses on the theory method and application of visual saliency detection:(1)The human visual system and some existing visual saliency detection models are discussed.Through the analysis of several typical saliency detection algorithms and the performance of each algorithm,we learned that the selection of image features and the choice of calculation strategies are important for the calculation speed and precision of detection result.And some of the saliency algorithms are used as the reference of the algorithm proposed in this paper.(2)In order to improve the accuracy and computational speed of saliency detection based on contrast,a context-contrast saliency detection method by introducing color,position and frequency prior knowledge was proposed.The algorithm uses the idea of local and global contrast in the context saliency detection method,uses the super-pixel as basic unit of image processing to reduce the complexity of the algorithm by the preprocessing operation of image super-pixel segmentation;Then,the priori information contains frequency characteristics,color characteristics and location feature is incorporated into the context contrast detection to produce a more accurate full resolution saliency map.(3)In order to demonstrate feasibility and validity of the proposed algorithm,It has been compared with several typical saliency detection algorithms on international public available database by using several commonly used significance test indexes.The experimental results show the proposed method has good feasibility and validity in saliency detection.In addition,this method is applied in the field of image salient target extraction.