Research On No-reference Qualityassessment For Distorted Images

During the process of image, it will cause the distortion of image for various reasons,such as incomplete imaging system,faulty transmission medium, imperfect storage devices,the relative motion of the object,noise pollution,etc.so it is necessary to establish a imagequality assessment system in the related fields of image processing. The objective evaluationmethod is divided into the full reference quality assessment method, the reduced referencequality assessment method and the no reference quality assessment method.The former twokinds of these methods need the information of the original image to evaluate the qualitydistorted image. The no reference quality assessment method is only to use the information ofdistorted images to evaluate the quality of images without any information of the originalimages.So it is called"blind evaluation".In the practical application of image qualityassessment, it is difficult to obtain information of the original image in many cases, so therehas a great significance of research the no reference quality assessment method.Insight into the research value,the theoretical basis and related some typical algorithms ofno reference quality assessment,there is a discovery that many no reference image qualityassessment methods cannot give better assessment results for blurred images.Considering thisphenomenon, the paper proposes a no-reference image quality assessment method based onthe texture and structure features of blurred image.The texture features represent the changeof image details and the structure features represent the change of the whole imageinformation.So this paper measures the distortion of blurred images from the twoaspects.Experiments indicate that the prediction results show stability and little differencefrom the experimental data.And the paper has a better prediction results compared with thepeak signal-to-noise ratio method,the structural similarity method and the no-referencestructural sharpness method.Considering that many no reference image quality assessment methods cannot directlygive assessment results for color images,and it is firstly to obtain color images not gray scale images in the practical application of image quality assessment.These methods ignore theircolor characteristics of images to evaluate the quality of their gray scale images,it will causedeviation in the prediction results for some evaluation algorithms that the color characteristichas a serious distortion.For this situation, research the several kinds of color model ofimages and some typical color image quality assessment algorithm,this paper proposes ano-reference image quality assessment method for color images using discrete quaternionFourier transforms.It makes parallel processing of RGB channels so that can make full use ofthe whole information of the color image. the prediction result has a good consistency withthe human visual perception.