| Image noise is one of the factors that hinder people to receive true information of images, and it is generated during transmitting images. Complex noise is associated with images produced by the Charge-coupled Device (CCD) and environmental factors also lead to noise. Image denoising, which is able to reduce noise, is a basic issue in the area of image processing.Noise generated during transmission is the main cause of image degradation and noise is introduced for many reasons. The proper denoising methods applied on different noisy environment are not same. In terms of denoising methods, there are two mainstream orientations, denoisng in special domain and transform domain. For the demand of high-quality images, more and more denoising methods started to use both of these two theories to denoise. For instance, BM3D mixes theory in special domain and transform domain to denoise. It consists of block-matching and filtering in transform domain:Block-matching groups the similar blocks and filtering in frequency domain releases the denoising process in transform domain. The block-matching with 3D transform domain collaborative filtering (BM3D) has excellent denoising performance so it has already become one of the mainstream demoising algorithms. However, the complex process limits its application in embedded systems, especially the systems which only support integer operators.Considering these shortages, in this paper, we propose an integer BM3D method, which integerizes all the floating point operators in BM3D, including 2D discrete wavelet transform (DWT), discrete cosine transform (DCT), Wiener filtering, ID hadamard transform and Kaiser window function in aggregation. The integer transform and filtering improve the efficiency of this denoising method and make it possible to apply the method to embedded system. In experiments, the proposed method produces comparable denoising results with the original BM3D, even better results when the noise level is high. Furthermore, the computing complexity is reduced almost 20%. Integer processes enable it to be transplanted into embedded systems. The integer transform and filtering boost the efficiency of the modified denoising method. |
