Research On Intelligent Image Recognition Technology

During image acquisition, transmission will produce different levels of noise, which affects the accuracy of image recognition. Therefore, before an image recognition for image filtering it is particularly important. Image includes two types of noise :Gaussian noise and salt and pepper noise, it proposes two noise filtering algorithm against two characteristics.According to Gaussian noise characteristics and the defects of the traditional filter algorithm on the image detail protection, an improved fuzzy weighted mean filter algorithm based on image pixel level of noise was proposed in this paper. At first, the algorithm based on image pixel level of noise gets the first time filtering image and the original image estimated histogram. According to the histogram determine the fuzzy memership function, the fuzzy weighted mean filter is used to filter those pixels in the first time filtering image whose gray scale values are less than 25.For salt and pepper an improved median filtering algorithm based on support vector machine was proposed in this paper. At first,the algorithm filters the noise image by using median filtering and then defuzzification operations are performed on the filtered image. For those pixels in the defuzzified image whose gray scale values are the maximum or minimum value,support vector machine(SVM) classification are used to decide whether they are noise points.Finally support vector machine(SVM) regression is used to recover original signals from the noise points. For two kinds of algorithms for simulation experiment and simulation analyses show that our algorithm can effectively remove salt and pepper noise, and has higher peak signal to noise ratio.In image recognition, select probabilistic neural network as a classifier, but when the training sample increases, probabilistic neural network operating speed is reduced. Then proposed a dimension reduction probability neural network recognition algorithm, the algorithm based on wavelet decomposition and Fast ICA algorithm to extract image feature vector, and then use probability neural network to get the feature vector for recognition, greatly improving the accuracy and speed. The experimental results show that the proposed dimension reduction probability neural network algorithm to achieve the rapid and high precision image recognition effect.