And Image Quality Evaluation Method Based On Local Feature Noise Estimate

In the generation and transmission process of the digital image, due to the influence of various factors like the imaging system and the communication channel, etc., there will generated different levels noise. The noise will not only affect the image quality and visual effects, but also interfere with the subsequent various image processing such as feature extraction and target detection. Image quality assessment in field of image processing, particular in image enhancement, played an important role.Image denoising will improve the performance of subsequent image analysis algorithm.Many denoising methods will utilize the priori knowledge of the noise, such as noise type, intensity and environment to design the corresponding filter. However, these priori knowledge in the image processing methods obtained by subjective conjecture or assumptions instead of objective description of the image noise. In this case, the image denoising processing is often blindness. Therefore, a reasonably and accurate estimate of the image noise is essential. In most image noise estimation algorithm, the noise signal is usually assumed to be independent and identically distributed additive the fixed zero mean Gaussian noise. So the purpose of noise estimation is to estimate the noise variance σn2from the noisy image.In the process of image processing, due to the noise and the influence of various factors, image quality problems is increasingly prominent, image quality assessment is emerged. So understand and evaluate the image quality plays an important role in many fields. In the image processing system, using the image quality assessment can timely adjust the parameters, so that the system has been optimized and improved. In practical applications, image quality in the processing have uneven grade. The accurately assessment of image quality, and then combining the results of assessment for further processing, has a very important significance in the application.This paper deeply studies the digital image noise estimation method and the image quality assessment method. On the basis of review and summarize previous work and combined with the practical application,this paper presents methods of image noise estimation algorithm and image quality assessment based on the image local structural features. The basic theoretical of these two methods are the degree of image feature. Noise variance estimation method based on statistical hypothesis tests presents a fast and reliable noise estimation algorithm for additive white Gaussian noise. Firstly,the proposed algorithm provides a way to measure the degree of image feature based on statistical hypothesis tests (SHT).The proposed algorithm distinguishes homogeneous blocks and non-homogeneous blocks by the degree of image feature. Secondly, sets the minimal variance of these homogeneous blocks as a reference variance. And then finds more homogeneous blocks whose variances are similar to the reference variance and which are not contain edge. Lastly, the noise variance is estimated from these homogeneous blocks by a weighted averaging process according to the degree of image feature. Compared with the existing noise estimation methods, the proposed algorithm performs well in the estimation precision and suitable for the Gaussian noise-infected images. The image quality assessment method based on the image local structural features take the image distortion caused by noise and fuzzy into account, use local statistical method to extract local structural features of each pixel, and then calculate the quality factor to assess image quality. Firstly, we calculate the maximum directional feature value Y5and minimum directional feature value Y1of each pixel in the neighborhood of5×5, and then calculate the local quality metric qi for each pixel, finally we calculate the entire image quality assessment result Q=Σqi,. This result can assess the image quality.After detailed account for the basic principle and specific methods of the proposed algorithm, this paper also performed a large number of experimental simulation in MATLAB7.0. The simulation compared our methods with exsiting methods, the comparison result shows the superiority of the proposed methods.