Microscopic texture components classification for image segmentation

The paper presents the possibility of using a sliding window for image feature extraction in order to identify image regions of interest. The study includes the comparison of feature extraction methods both in the space and frequency domains using the discrete Fourier transform and the discrete wavelet transform to achieve reliable classification results for a given application. The compactness of feature clusters is evaluated exploiting a proposed numerical criterion. In case of real image data, the clusters compactness can often be improved by employing a chosen smoothing method on the raw data. In this paper, the procedure of smoothing, feature extraction and classification is applied to microscopic images of aluminum alloys in order to isolate regions of similar properties and to study their relationship. To achieve this goal the sliding window classification results are combined and isolated misclassified subregions repaired. The proportion of misclassified regions is then used for the evaluation of the efficiency of the proposed method along with the proposed measure of cluster compactness.