CAJ | 학술논문

针对目前广泛使用的Non-local Mean和Block-matching and 3D Filtering等方法虽然在高斯噪声消除方面效果不错,但均采用了块匹配操作,从而导致非常大的计算量以及较长的时间消耗,很难满足短时间内大批量快速图像处理的需要。对此,提出一类修正的巴特沃斯函数,在频率域上对图像进行快速降噪。数值结果表明该算法对于中小强度的噪声干扰降噪效果良好,能够保持图像的主要细节。与BM3D,全变差正则化和NLM法相比,其峰值信噪比和全变差正则化法以及NLM差别不大,但耗时却只有BM3D和全变差正则化法的1/50,比NLM更是快了近1000倍,能够适应大批量的图像处理需要。
침대목전엄범사용적Non-local Mean화Block-matching and 3D Filtering등방법수연재고사조성소제방면효과불착,단균채용료괴필배조작,종이도치비상대적계산량이급교장적시간소모,흔난만족단시간내대비량쾌속도상처리적수요。대차,제출일류수정적파특옥사함수,재빈솔역상대도상진행쾌속강조。수치결과표명해산법대우중소강도적조성간우강조효과량호,능구보지도상적주요세절。여BM3D,전변차정칙화화NLM법상비,기봉치신조비화전변차정칙화법이급NLM차별불대,단모시각지유BM3D화전변차정칙화법적1/50,비NLM경시쾌료근1000배,능구괄응대비량적도상처리수요。
A fast method to reduce additive Gaussian noise in images was discussed. Some methods, like Non-local Mean(NLM) and BM3D, perform very well in noise suppression but in them the block matching is carried out, which results in huge computation and long running time. So they are not competent for work that has to be finished in real time. In order to solve this problem, a kind of revised Butterworth function was proposed to remove additive Gaussian noise in images through frequency domain. The numerical results showed that the output of this method could preserve structures well in images after denoising when the noise magnitude was not very big. Compared with the three methods, i.e. BM3D, total variation regularization and NLM, the Peak Signal to Noise Ratio (PSNR) of the method was just a little smaller than Total variation regularization and NLM. However, the running time it takes was only one fifth of them and was 1 000 times faster than NLM. So it is suitable for huge number of image processings.