Research On The Methods Of Image Feature Extraction Based On Wavelet Transform

In the field of image processing and computer version, image recognition is a complicated and critical technology, but also one of the most critical issues is the image feature extraction. The image feature is the most basic inner feature used to distinguish other images. The image feature extraction separates from computer version and image processing, using computer to analysis and process image information, and then confirming the invariant image feature. This paper has deeply researched the computer image preprocess technology and image feature extraction methods based on wavelet transform.During the image feature extraction process, the image preprocess is very important, and it is very necessary to eliminate the image noise and improve the image quality. This paper uses the histogram equalization method based on wavelet transform to improve the image quality, and uses wavelet threshold denoising based on wavelet transform to eliminate the image noise. The traditional histogram equalization could lose some important image detail information; this will exert a bad profound influence on image feature extraction. In this paper, it proposes an improved method that combines the histogram equalization and wavelet transform. It extracts the low frequency image, using histogram to enhance the low frequency image. In addition, the high frequency image includes the major part of the image noise, thus using wavelet threshold denoising technique to reduce the noise in high frequency image. Finally using wavelet transform to reconstitute the low frequency image and high frequency image. So by using this improved method, it can not only enhance the image quality, but also reduce the noise in the original image. The experiment shows this improved method can actually enhance the image quality and improve the image feature extraction.This paper also has deeply researched the current method of texture feature extraction based on Gabor wavelet. In this paper, it uses mean and variance of Gabor wavelet parameters to create texture feature vector. However the texture feature extracted by Gabor wavelet doesn’t have the character of rotational invariance. This limites development and application in texture feature extraction field. This paper proposes an improved method; it circularly shifts elements in the texture feature vector extracted by Gabor wavelet. This method can make the vector has the character of rotational invariance. By using Euclidean distance method measures the distinction of before-rotation vector and after-rotation vector. The experiment shows that this method can make the texture feature vector has the rotational invariance character.In addition, this paper applies the above two improved methods into the UT-2000 IV Microcomputer Anti-misoperation System, that makes a very good effect, making the operation interface more friendly.