Research On Image Objective Quality Assessment

The uppermost way which human beings learn from is visual information that is received by human perceptional system,and among the visual information photos are the matters of primary importance.With the development of PC,digital communication,multimedia and network technology,as one of the most important informational intermedium,digital image/video has penetrated into individual's daily life and popularized among people.In the orocedures of image processing system,e.g., acquisition,processing,coding,storage,transmission and reproduction,digital image may be in degradation in visual quality,so evaluating image quality becomes a fundamental job in the field of image processing and computer vision.The objective of this thesis is to research on the structural distortion of the image,and further design the algorithms of objective assessment to evalue the image quality automatically and accurately by using the characteristics of structural distortion.In this thesis,we explore which consists of the novel idea about the expression and measurement of image structural importance,the image's projection energy and directional projection.The main work and innovations are listed as follows:(1)Introduce matching pursuit into image QA,and it is modified for the image QA applications.Matching pursuit projects the image on the basis set and extracts the image structure under the rule of the largest energy,and it is used widely in image and video coding.We modify matching pursuit to meet the image QA applications,and decompose and extract the important structural information from the reference image.(2)Propose an approach of expression of image structural information and measurement of structural importance.Human visual perception is hilly adaptive for extracting structural information from a scene.Generally according to the differences of the characteristic which the structure represents,the different structural information is always of different importance.On such an assumption,modified matching pursuit is introduced into extracting the structural information from images,charactering the structural information and measuring its importance. (3)Propose an image QA approach based on the importance of structural information.The proposed idea of structural importance is introduced into image QA.By extracting the structural information from the reference image, the different structural information and its importance measurement is gained.The structural distortion and quality predictive score are calculated by comparing the difference between the structural information of reference images and distortion ones.(4)Propose an image QA approach based on projection energy.On the assumption that any image distortion could be modeled as the difference of the PEVs(projection energy vector)between the reference image and the distortion one,an image QA approach is proposed based on signal projection energy.This approach uses the simple projection operator and simple parameters and it achieves high efficiency and low computational complexity,so it is adaptive to the real-time applications.This approach is of simple parameters and easy to calculate,thus it expects to work as an expansion for those mathematically defined ones for example PSNR.(5)Propose an image QA approach based directional projection.The image structure contains many pixels,edges and shape with directional characteristic mainly,and the directional characteristic is charactered by the images' projection on different direction.The distortion of the structural information could be modeled as the alteration of the directional characteristic.This approach is implemented by Radon transform which calculates the directional projection of images and develops the DPV (directional projection-based vector)to represent the directional characteristic.The structural distortion is calculated by comparing the difference of the DPVs between reference images and distortion ones.