No-Reference Image Quality Evaluation Method Study

With the development of computer and multimedia, digital images as an important information carrier, has penetrated into people’s learning life.Image distortion exists in each processing link.Hence the need for a mathematical model or method to measure the quality of the image, to the image play a reasonable score. This model or image quality evaluation method is IQA(Image Quality Assessment).In order to evaluate the distortion of the image quality, this paper analyze the nature and characteristics of the image transform domain, combined with the advantages of neural network modeling,this paper proposes a neural network model optimization based on contourlet domain and cuckoo. In this model, the image sequence after the extraction of energy contourlet transform transform,followed by neural networks for energy information without reference image quality assessment. Due to generalized neural network has "characteristics shield" in nature, so the Cuckoo search for the neural network optimization.To measure the degree of degradation of the image, through the study of Curvelet transform found that, compared to the contour of the wavelet transform to better capture images of their geometric features.So put a no-reference image quality assessment method based on Curvelet transform and neural networks. Firstly, the energy eigenvalues extracted Curvelet coefficient, then the use of asymmetric generalized Gaussian fit, and find Curvelet direction information, get a set of feature evaluation value. Feature vectors are then fed into the cuckoo optimized neural network training, theestablishment of nonlinear relationship characteristic statistical values and subjective evaluation value, and then get the predicted value of the input image with the mappings. With the LIVE database to validate the algorithm, experimental results show that compared with the methods of the current literature, the algorithm has a higher consistency and accuracy.