Image Noise Removal Method Based On Multiscale Interpolation Framework And Its Application Research

Image is the main source for human to acquire and exchange information.According to the statistics,more than 80% information accepted by human is expressed by the image,so image plays an important role in people's daily life and work.However,images are usually contaminated by various types of noise during acquisition,transmission and recording processes because of bad sensors and transmission channel interference.This leads to low definition and poor quality of the acquired images,which seriously affects the visual effect and even hinders people from obtaining useful information.Therefore,how to effectively recover clear images from a large amount of noise has always been a classic issue,and it has important practical significance and application value for improving the image quality.Impulse noise is one commonly encountered noise type.Among the uncertainties involved in impulse noise,the randomness and fuzziness are the two most important features in corrupted images.Generally speaking,the randomness mainly refers to the fact that the pixels is randomly contaminated by noise and the noise pixels are randomly set to the maximum or minimum value.The fuzziness mainly focuses on the pixels with the extreme values whether they belong to the noise or not.Aiming at the shortcomings and the insufficiency in the existing denoising methods,this paper proposes a new framework to remove the impulse noise,which obtains good denoising effect.The work in this paper mainly includes the following aspects:(1)Several existing denoising methods based on impulse noise detection are analyzed in detail,and the shortcomings and the insufficiency in these methods are summarized as follows.First,all pixels are regarded as an individual.Each pixel is estimated independently rather than jointly estimating pixels in neighborhood,which ignores the correlation between pixels.Second,these methods only make full use of the statistical information while completely ignoring the structure information during removing the image impulse noise.(2)This paper discusses the application of autoregressive model in image processing,and focuses on introducing the autoregressive model into the image impulse noise removal processing which converts the image denoising problems to the interpolation process.The experimental result verifies that the autoregressive model applied in the impulse noise suppression is feasible.(3)The multiscale estimation method and the idea of autoregressive model are introduced into the impulse noise suppression,which leads to a new impulse noise processing method based on multi-scale interpolation.The whole idea of denoising includes the following parts: First,it uses the histogram method to detect the noise pixels and make the noise pixels set to be the missing pixels;Second,these missing pixels are pre-filtered by using the switching median filter;Finally,more low resolution sub-images can be obtained after continuously down-sampling the pre-filtered image.Combining the statistical features with structure characteristics of each sub-image,it uses autoregressive model to multiscale interpolate each sub-image until getting more full resolution images.Then it sums these full resolution images in different phase to recover clear images while suppressing impulse noise.In this method,all noise-free pixels are used as known constraints in the whole process of noise suppression.The experimental simulation verifies that the proposed method produces better results than the existing methods in subjective visual and objective evaluation.It can be used to remove the high density impulse noise while preserving image details and structural information.