On Image Noise Recognition Based On SVM And Wavelet-transform

Digital image signals may produce different types of noise for being affected by various random disturbances when in the process of being collected or transmuted. The analysis of the noised images will become difficult, so it is a critical issue to denoise the image. Relevant study has been going on for decades. The key step in restoring the image affected by impulse noise is to identify the polluted pixels accurately. Having Analyzed and synthesed the recent achievements on denoising the impulse in image, this thesis proposed a method based on Mean of Indicating Vector Per Unit (MIVP) for restoring the image effected by impulse noise.The main work is as follows:Firstly, an impulse noise filter based on combination wavelet transform with support vector machines classifier SVC is analyzed after briefly introducing the wavelet transform and support vector machines. Analysis revealed that there is a flaw in the filter based on combination wavelet transform with SVC.The flaw is that such filter has a higher miss-judging rate of identifying the pixel polluted by impulse noise as the one not polluted, and that the filtering effectiveness depends on SVC's training samples. Furthermore, experiments show that SVC and the wavelet transform are not complementary each other in judging the polluted pixels.Secondly, a Mean-of-Indicating-Vector-Per-Unit based method is proposed for identifying impulse noise, according to both the law of large numbers and law of central limits in statistical theory, the idea is proposed based on Mean of Indicating Vector PerUnit(MIVP) for detect the pixels polluted by impulse noise. And theories and empirical two sides to give a demonstration of the algorithm mivp the legitimacy of and superiority. Finally, The impulse noise filter is designed based on the combinationMean-of-Indicating-Vector-Per-Unit method with wavelet transform method,The results shows that Mean-of-Indicating-Vector-Per-Unit based method is feasible for identifying impulse noise and can be put into practical engineering use after being proper improved.