Research On Image Stitching Quality Assessment Based On Local Phase Congruency Features

Image stitching has been widely used in VR,medical,video surveillance and other fields,but the research on the quality assessment of stitched images has not kept pace with the development of image stitching.This hinders the development of automated evaluation of image stitching projects.The quality assessment of image stitching is different from it of the traditional image.The noise and distortion of traditional image quality assessment are mostly global uniform distribution,but the quality of stitched images is more affected by local artifacts.Therefore,the traditional quality evaluation algorithm will lead to some errors.The research in this paper is divided into two parts: local phase consistency feature design for image stitching,and quality assessment design with local phase consistency features.The dislocation distortion caused by image stitching is analyzed in detail,and some performances of HVS are applied to improve the performance of the image stitching quality evaluation algorithm.The main work of this paper can be summarized as the following three points.Firstly,with the image stitching method based on feature points,different mosaic dislocation images are obtained by simulating different camera positions and postures.Building a stitching database based on APAP.Testing and analyzing multiple mainstream image quality assessment algorithms on this database.Secondly,based on the phase consistency algorithm,a locally normalized grid oriented gradient histogram is designed.The feature information of local image can be described with this method.This histogram with local area normalization will effectively reduce the dimension of feature description and suppress the influence of noise and light variation energy.Then,the local features of the misplaced image are extracted and compared with the features of the original image.The results show that it can represent and reflect certain misplaced information.Finally,an algorithm of stitching image quality evaluation based on local features is proposed.After the local features of the image are encoded,they are put into the minimum pool for screening.The proportional coefficient of the screening filter is determined by JND threshold.The application of its minimum pool and JND threshold coefficient is derived from some features of HVS.The test was carried out on the stitching database based on APAP.The results show that the proposed algorithm can distinguish the images with different degrees of dislocation and get the results close to the subjective evaluation.It is found that the SROCC and PLCC with the subjective evaluation results can reach 0.87 and 0.86.This paper studies the quality evaluation of image stitching results.A local feature based on phase consistency and a quality evaluation algorithm based on local phase consistency feature are proposed.Our algorithm can describe the characteristics of different degree of dislocation effectively,and get good objective evaluation results for the mosaic images with different degree of dislocation.