The Research On No Reference Image Quality Assessment Alogrithm

About three-quarter of human information is extracted from images.With the development of society and technology.people’s life hase been riched. All kinds of image products into daily life.High-definition TV, digital camera,3d images such as electronic products has become a common phenomenon. In the process of acquisition, displaying, coding, recording, processing and transmission, images will be inevitably suffered from distortion with different extents and different kinds. Therefore, design a reasonable algorithm of image quality evaluation is helpful to evaluate digital image products.Image quality assessment is mainly divided into subjective evaluation and objective evaluation. Because of the subjective evaluation is influenced by human emotion, randomness, not repetitive. Therefore, the objective evaluation has being a research focus in the field.Designing the objective image quality assessment method more consistent with HVS is most important in research.Objective image quality assessment method is divided into Full-Reference, Reduced-Reference and No-Reference. In many practical applications, there is usually no original reference images or is unable to get the original reference image.Because No-Reference assessment don’t need reference image,it has a broad application prospects.This article mainly aims at the study on no reference image quality assessment. The main work and innovations are listed as follows:1.Now there are some good effect for single type of distortion without reference image quality assessment methods.If the type of the image distortion can be classified accurately, then we can use the existing for one particular type of distortion of image quality evaluation method for quality evaluation. Based on this thought, this paper puts forward a kind of image distortion type classification method based on wavelet transform. First, by wavelet transform for image feature extraction, feature vector. Then use support vector machine (SVM) to study feature vector. By using the particle swarm algorithm to the optimization of the parameters of support vector machine (SVM), to improve the distortion image classification accuracy. Through do the test in the LIVE2database and shawn qu database, verify the rationality and effectiveness of this method.2.Based on the classification of distortion images,this paper proposed a No-Reference image quality assessment method. The distortion of image wavelet transform in the first place, then use support vector machine (SVM) to classify the image. According to classification of different types, choose specific quality evaluation method for quality evaluation. Among them, for JPEG distortion type with proposed by Wang Z a against JPEG compression without reference image quality evaluation method for the evaluation; Blur and noise distortion of these two types are selected based on gradient similarity method for the evaluation; For FF, WN, these two types of distortion, is the support vector machine (SVM) regression method was adopted to estimate its quality. Finally, the image is obtained by weighting the quality of the final value. Simulation experiments on the LIVE2database of this method, the evaluation results objective evaluation and subjective results have good consistency.3.Proposed a no-reference stereoscopic image quality assessment.Core idea of this method is a kind of no reference plane image quality evaluation method and in combination with the characteristics of stereo image, the stereoscopic image quality evaluation. Here, use NIQE of image quality evaluation first, and then by calculating the three-dimensional image of left and right view parallax figure, reoccupy NIQE directly to evaluate quality of parallax figure, the final weighted three-dimensional images of the final quality.The method on LIVE3database evaluation results and subjective evaluation results have a good consistency.