The Study On Stereoscopic Image Quality Assessment Based On Eye-Tracking

In recent years,Stereoscopic 3D image(S3D)technology has been widely used in various industries such as film and television,multimedia and so on because it can reconstruct stereo depth information and bring more natural immersive visual experience.However,the problems such as visual fatigue,sourness of eyes and low comfort,which occur when viewing 3D images,have become an important factor restricting the popularization and application of 3D images.The planar stereo image is composed of two plane images in the left and right eyes of the human eye,and is processed by the vision system to make the human eye perceive the depth information of the image and establish stereoscopic vision.With the deepening of the human visual system,more and more researchers began to combine the human visual principle,to study the 3D image discomfort problem.Eye Tracker technology can record the whole process of human eye movement when viewing 3D images.It is a commonly used data representation in human visual research.It is the main content of this paper to extract the comfort-related features from the eye tracking data and study the 3D image quality assessment method based on eye movement data.In this paper,we first develop a Web-based 3D image quality assessment platform.This platform is based on the Web technology,carries on the program control to the 3D image quality evaluation experiment and reliably collects the evaluation data,and provides the unified flow control and the management for the eye movement experiment.Based on the platform,we designed and implemented the subjective experiments,and established the SIED image library needed for this study.During the experiment,the platform is stable,the operation is simple and efficient,and the data is accurate,which greatly reduces the cost of subjective experiment.In addition,we extracted the SSF(Simple Statistical Features),SDF(Statistical Distribution Features),and NRF(Neural Response Features),which are the three comfort features of IEEE-SA and EPFL And the 3D image quality evaluation model based on five different regression models was established.The performance of the model was analyzed,which provided the reference for the selection of regression model and the analysis of model performance in the establishment of 3D image quality evaluation model.Subsequently,the 3D image quality evaluation under the interest mode is studied in this paper.This paper analyzes the accuracy and precision of eye tracking data in 3D environment,and studied the method of using ROI to mark the region of interest.The region of interest(ROI)was extracted by kernel density estimation using the eye fixation data,and the correlation between the parallax of the image region of interest and the image MOS was analyzed.The experiential formula of the target parallax of interest and the MOS was obtained to provide the3 D content security and comfort enhancement.In addition,this paper extracts the comfortrelated features from the ROIs,and builds a 3D image quality evaluation model based on eyemoving visual attention.Compared with SSF,SDF and NRF,the quality evaluation model based on region of interest(ROI)has improved the prediction performance.Finally,eye tracking data was used to analyze the eye movement characteristics related to3 D image quality.In addition to using the gaze data to mark the region of interest,the eye tracking data itself can characterize the eye movement process when viewing 3D images,especially the vergence adjustment process closely related to the human eye stereo vision mechanism.In this paper,we describe the eye movement data of human vision under stereo vision.Combining the Foveation region and Panum fusion region in the human eye vision,we extract the FRF(Foveation Region)related to the human eye viewing stereoscopic image Features and PFF(Panum's Fusion Features)features.Through the establishment of 3D image quality evaluation model based on eye movement process,the relationship between the eye movement characteristics and image comfort was analyzed.Experimental results,FRF and PFF these two groups of eye movement characteristics and image quality are closely related.This study shows that the use of eye movement data can help analyze the human eye vision,mining eye movement characteristics reflect image quality,eye tracking data in the field of 3D image quality evaluation has important research value.