Research And Implementaiton Of The Image Quality Assessment Method

With the rapid development of the Internet，the use and requirements of the digitalimage as an information carriers is increasing higher in order to meet the growingabundant modern business. The fast-growing of the computer network greatly expandsthe application field of the image. The process of the image acquisition, processing,transmission, storage, and display, etc. will produce different types and also differentdegrees of distortion. This will lead to an aberration of human’s information. Therefore,the establishment of relevant mathematical model to evaluate the image quality obtainedhas great significance in the field of image processing and application.This thesis first presents an image quality assessment method with statisticalcharacteristics based on linear regression. Firstly, on the basis of existing image qualityassessment methods, the thesis gives emphasis on four most widely used methods, theMean Square Error method (MSE) and the Power Signal-to-Noise Ratio (PSNR)method, the Structural Similarity Image Measure method (SSIM) and the informationfidelity method (VIF). Secondly, through the use of the Different Mean Objective Score(DMOS) provided by the LIVE Image Quality Assessment Database, three correlationcoefficients are used to measure the relevance between the four methods with theDMOS. So we can get a statistical conclusion about better methods. Then, we present amethod which is based on linear regression to normalize the existing assessmentmethods. We get an image quality assessment method based on ternary linear regression,whose English abbreviation is TLRE. Finally, the new assessment method proved moreeffective using the LIVE database.In addition, the thesis also proposes an assessment method based on image features.The target of the method is to evaluate the quality of different sized images. Wecalculate three low-level similarity coefficients of the color distribution, the shapecharacteristics and the distribution of the texture. Then three coefficients are normalizedto one and the image quality will be obtained. Through lots of experiments, this methodhas a very good effect on the same sized and different sized image quality assessment.