Research On Quality Assessment Of Panoramic Image By Multi-mosaic

Panoramic images can provide full view images in same scene for users.It has been widely used in various fields such as remote sensing technology,virtual reality.However,compared with the rapid development of image mosaic algorithms in the past ten years,there are insufficient and relatively lagging researches on the evaluation of stitching image quality.Different from the traditional image quality evaluation,the stitching image is the stitching of multiple images,the distortion exists locally,and the exact corresponding reference image cannot be found in the real scene.Therefore,the traditional quality evaluation algorithm is not suitable for the quality evaluation of stitched image.By analyzing the dislocation distortion caused by image stitching,taking the corrected unstitching image information as benchmark,using the principle of the best stitching algorithm and some characteristics of the Human Visual System,modeling local distortion of stitched images,this paper proposes a multi-feature fusion quality evaluation algorithm.The main research work is as follows:(1)By changing the Euler angle of the camera,different degrees of images are captured in same scene.And then the image stitching method based on the best seam are used to calculate dislocation stitching results.The dislocation distortion of these stitching images are deeply analyzed.(2)This paper designs a multi-feature fusion quality evaluation algorithm.Multi-scale and multi-directional subband information is obtained and the structural similarity calculation is performed through the controllable pyramid decomposition,which can well describe the structural characteristics of the image.Then phase consistency,gradient,and chroma similarity features are extracted,which are used to characterize different feature attributes of the image.Second,the feature pooling strategy is used to process each feature and combine it into a set of similar feature vectors for image quality detection device.After that,SVR support vector regression is used to conduct training tests,feature vectors is mapped to objective quality scores,and finally experimental analysis and comparison is performed on five standard IQA databases.The results show that this method can effectively describe the distortion of images and has better subjective consistency compared with the currently popular FRIQA method.(3)This paper proposes a stitching image quality algorithm based on the distortion area.For the same reference image that cannot be obtained,the two corrected images to be stitched are selected as the reference,and combine with the common overlapping area of three images.Two matrices of encoding similarity are generated by decoding the similarity of two overlapping areas,which created by the two images and the stitching image overlap respectively,and then mapping to the overlap image by compare with.The JND threshold is used to filter the image blocks of the stitching area.The quality evaluation score is calculated in this area.The test,which carried out on a database,shows that the algorithm can distinguish images with different dislocations,and obtain results close to subjective scores.