Research On Image Denoising Algorithm Based On Improved 3D Block Matched Filtering

The development of society drives the high-speed operation of information and data.In order to obtain the information we need more intuitively and vividly,digital images have entered the"mainstream"of information carriers.And with the continuous advancement of scientific and technological development,digital images have played a critical role in education,medical care,communications,smart agriculture,and communications in different forms.The rapid development has also brought certain drawbacks,especially the problem of image clarity,which hinders the acquisition of real information to a certain extent.In the process of image acquisition and transmission,due to the influence of the external environment and the sensor itself,the image will inevitably be affected by noise,resulting in a decrease in quality,and its accuracy,intuitiveness,efficiency and practicality will be greatly reduced.^[1].Therefore,image denoising technology came into being,which attracted many researchers.In the process of improvement,the three-dimensional block-matching and 3D Filtering(BM3D)algorithm has achieved many affirmations due to its unique and effective advantages.As one of the best denoising methods in the field of digital images,there are still some shortcomings:low computational efficiency,superposition of high computational cost,and insignificant denoising effect of edge texture information.These shortcomings limit t he algorithm's performance.widely used.In order to obtain a clearer image,this article proposes an improvement to the original BM3D algorithm.The main innovative work and research results of the thesis are as follows:(1)The BM3D algorithm shows its powerful image denoising ability,which is achieved by block matching,filtering and aggregation of the three-dimensional array generated by the noise image.In the paper,firstly,the grouping process is optimized for the time-consuming problem of searchi ng for similar blocks.Secondly,it is proposed to reduce the high computational cost by using an adaptive algorithm based on the pre-classification of the difference coefficient.After pre-classification,two block subsets with different local structure i nformation are obtained,which are a complex texture area and a flat area.Therefore,different matching methods are used for different regions,which will better eliminate noise at a lower computational cost.Experimental results show that compared with t he original BM3D algorithm,its computational cost is significantly reduced.(2)In the BM3D algorithm,because the complex texture area does not have enough similar blocks,the edge denoising effect is not obvious.For this area,the original method canno t recover as much complex texture information as possible.This paper proposes a method that combines the improved TV model and BM3D to restore the image for this area.Compared with the basic ROFTV model,the improved model has made a certain improvement in subjective and objective evaluation,laying a good foundation for perfecting details such as edge information.In the course of the experiment,we found that there are still some blurring problems on the edges of complex texture areas.Therefore,this ar ea uses an improved Prewitt operator edge detection algorithm for edge detection and enhancement,Further improve the experimental effect.It can be said that the effective combination of the two improved algorithms makes up for the main shortcomings of th e original BM3D algorithm.Regardless of the large amount of calculation or the difficulty of denoising in complex texture areas,the effect of image restoration experiments has been improved.