A proposed method of image recognition using a recursive matrix generalized from wavelet coefficients

摘要

In the last ten years the wavelet transform has diffused in most digital signal processing application. It plays an important role in image processing. This paper suggests new methods of image recognition that use the wavelet transform as rotation invariant feature tool with multi-resolution generality of the image. The idea here, is to create a unique matrix for each resolution image. It contains the orthogonal split features maintaining form this resolution projection of the image. This matrix which companion with this image is unique and refers to this image only, and it is named the recursive matrix. This paper also suggests the direct recognition method, which generate recursive family matrices. These matrices are a tree of all the recursive matrices of the reference images that are generated for every resolution. The recursive family matrix for any resolution contains the orthogonal features of all images under the same group. Two dimension convolutions are used to build recognition matrices. This strengthens the features that belong to both recursive matrix of the image and the recursive family matrix. These methods give good performance in image recognition, and showed a powerful rotation invariant tool in image processing.