Still Image Compression Method Base on Visual Attention Mechanism
|School||Hebei University of Technology|
|Course||Applied Computer Technology|
|Keywords||Perceptual Model Contour Feature Focus of Attention Image Compression|
The region of interest image compression method can without lose important information atthe same time and can effectively reduce the amount of data. In order to extract the main visualinformation from natural images automatically, the visual perception model was introduced inthis article. On the basis of the analysis of the bottom-up visual attention model driving by dataand the up-bottom visual attention model associated with tasks, this paper focused on Itti visualcomputing model. A new method was proposed by combining the idea of Itti model and thespecific application of image compression. This method provides a solution to the image thatregion of interest in the image is difficult to determine automatically. Compared with Itti model,the improvements model mainly include the following work:1. Excepting the features of color, lightness and direction, the improved model uses cannyoperators to extract the edge of the objects in the image and fills the image of containing theedge of objects. Then, contour feature of the filled image is fused into the model’s calculationprocess to attain total saliency map of the image and improves the accuracy of significantregional.2. In the model, the feature map merger strategy is improved. The nonlinear merger strategyreplaces the previous method of feature maps being added on average. Considering that thefeature map in which the density of significant points is too large doesn’t play an important rolein the merging process, this article removes these feature maps in which he density of s ignificantpoints is too large by setting corresponding threshold.3. This method improves the original visual attention focus shifting strategy. All thesignificant regional of input images are determined based on the total saliency maps only once.Therefore, it greatly reduces the running time. The execution time of the improved model is3%of the previous. Because of its high speed, the new method is more suitable for the accurateextraction of ROI in the image compression.Besides, JPEG2000image compression standard is discussed in this article. This articlefocuses mainly two methods, general translation method and maximum offset method. It givesthe specific method how to combine the improved perceptual model with the image compression. Firstly, the mask of ROI is determined by the perceptual model. Then, the mask and itscorresponding image are used to complete the image’s ROI compression. At last the newalgorithm in this paper is experimented and the experimental results proved the effectiveness ofthe algorithm.