Study On Saliency-based Subjective/Objective Classification And Assessment Methods For Stereoscopic Images

The stereoscopic display technology is popularized and applied in more and more fields because of its' visual immersive sense for users.Binocular stereo display,which is the most common type,can bring visual discomfort to users as a result of the inconsistency between binocular converge and accommodation.Most of the existing methods for visual discomfort evaluation based on the subjective questionnaire so that lack of reliable objective evaluation results.At the same time,for stereo image comfort classification,most of the present algorithms have low classification precision due to the shortage of reliable image features.To solve the above problems and improve the precision of evaluation and classification for stereo images comfort,this research has carried out the following innovative works:(1)Proposed a stereoscopic visual comfort evaluation method based on images' saliency difference.In order to realize the objective evaluation for stereoscopic display comfort,this paper presented an evaluation method that combined ECG signal and eye movement parameters by selecting stereo images with different salient area as the source of the experimental stimulus and synchronizing the ECG and eye movement parameters of the subjects during the experiment.The evaluation of stereo visual comfort was accomplished by combining the subjective questionnaire results.This method improve the precision of evaluation by combining the ECG signals and the eye movement parameters.(2)Proposed a multi-scale features fusion based on stereo images super pixel segmentation for saliency detection.The current saliency algorithms are limited to the poor salient area detection,feature maps' fusion and the lack of images' features.To solve these problems,this paper proposed an improved algorithm for detecting stereoscopic images' salient area.In this algorithm,super pixel method was used to segmente the original RGB images.After blended the multi-color-channel feature maps and depth map based on minimum barrier difference algorithm,the final saliency map was computed after add the original images' background cues.The experiment results showed that the proposed method could obtain a better salient area extraction effect.(3)Proposed a classification model for stereo image comfort based on image saliency features.In order to make up the deficiency of present studies,this paper proposed a support vector machine of Gauss Radial Basis kernel classification model.The model achieved the classification according to image saliency features.Except that,this research also based on weighted feature...