the human vision system is always looking for important and valuable areas to get the most information from the visual data in the shortest possible time. For this purpose, visual attention guides the vision system to...
详细信息
ISBN:
(纸本)9781538644058
the human vision system is always looking for important and valuable areas to get the most information from the visual data in the shortest possible time. For this purpose, visual attention guides the vision system to the salient regions. Unfortunately, some of the salient regions, in terms of the human vision system, have a high potential for attracting attention, but do not contain important and useful information. Therefore, attention and concentration on these salient regions will cause the loss of time and distraction of the audience's senses from the main subject of the image. Therefore, detecting and eliminating these regions, which are called distractors, can be of great help in order to increase the quality of the image, increase the accuracy of the algorithms based on saliency detection and not losing a lot of time in real-time applications. In this paper, pixel-based features are extracted from different approaches for training and test images and after segmentation the images, segment-based features are generated. After determining the class of each segment of the training images according to the corresponding masks, the test image segments are classified according to the TPTSSR method based on sparse coding and representation system in terms of the severity of the distractor in the 9 different classes. To evaluate MSE metric leads to unreliable results in conditions of class unbalancing, a new metric will be introduced to evaluate the results. The implemented results show that the proposed method has a higher accuracy than the previous ones in terms of these two metrics.
暂无评论