The3D Seismic Image Denoising Method Research Based On Trilateral Structure-oriented Filtering

With the rapid development of multimedia technology and the Internet technologyof the information age, images as the main form of information play an increasinglyimportant role in people’s learning, living and working. But images will be inevitablydegraded during the collection, transmission and storage, which bring a lot ofinconvenience to the practical application. Therefore, the urgent needs of imagedenoising technology in more and more areas are increasingly strong. It is required thatimage denoising can eliminate the noise and maintain the image texture detail. Due tothe band of degraded noise and part of the image texture is aliasing, smoothing noiseand keeping details are often a contradiction, and this is an important and challengingdirection and topics in the field of image processing.The paper firstly analyzes the image degradation model and common noise models,describes the advantages and disadvantages of several classic image denoising methods.Then around the regularization of mixed norm method and structure-oriented filteringmethod the paper does an in-depth analysis and research, and these two methods isapplied to a special class of images—three-dimensional seismic images. In oilexploration, through three-dimensional artificial excitation earthquake, a series ofthree-dimensional digital image (often referred to as a three-dimensional seismic data inthe field of oil exploration) which can reflect underground structures and rock featurescan be obtained, according to some special and unique problems of seismic image noisereduction, the paper develops a noise reduction algorithm applied to thethree-dimensional seismic images. The specific works are in the following two aspects:1.3D seismic image data acquisition environment is very complex, so widespreadpresence of Gaussian and non-Gaussian noise are in seismic images. This requires noisereduction algorithm to suppress both noise simultaneously. To solve this problem, thepaper put forward a seismic image noise reduction methods based on regularized mixednorm, the method uses the L2norm to suppress Gaussian and super-Gaussian noise, anduses the L4norm suppress the sub-Gaussian noise. The weights of L2norm and L4norm are adjusted based on the continuity factor of the image gradient structure, to optimize the performance of noise suppression adaptively.2. There is a variety of geological information in3D seismic image, and thisinformation cannot be smoothed out in noise reduction, therefore, it’s necessary foredge-preserving during the smoothing process. The paper raised trilateralstructure-oriented filtering noise reduction methods; the method selected the trilateralfilter function as the structure-oriented filtering kernel function. Compared withGaussian kernel, this method can be better to preserve the image texture detailinformation. From the experimental results of the actual data，it is verified that trilateralstructure-oriented filtering is better than Gaussian kernel structure-oriented filteringmethod.Finally, the proposed noise reduction algorithm is applied to the actual seismicwork area, and noise reduction results were compared with conventional noise reductionmethod. The results show that the proposed methods performance better than thetraditional noise reduction methods in the suppression of non-Gaussian noise andkeeping the edges texture information of the image.