Based on Fractal Dimension Calculation by the Area Cover for Extraction of Shift Invariant Multiwavelet Features of Texture Images

In this paper, we propose an effective method based on fractal dimension calculation by the area cover for extraction of shift invariant multiwavelet features of texture images. The feature extraction process involves a normalization followed by a shift invariant multiwavelet packet transform. The normalization converts a given image into a size invariant image which is then passed to the shift invariant multiwavelet packet transform to generate subbands of shift invariant wavelet coefficients. Then we convert the multiwavelet coefficients matrix to a smaller dimension correlation matrix. Then we employ a technique based on the fractal dimension (FD) and a new fractal dimension estimating method is proposed by taking the area instead of the volume covering in box-counting to estimate the FD. Three FD features are based on the original image, the above average/high gray level image, the below average/low gray level image. An energy signature is computed for each subband of these FD coefficients. In order to reduce feature dimensionality, only the most dominant wavelet energy signatures are selected as feature vector for classification.