Chainlet based Ear Recognition using Image Multi-Banding and Support Vector Machine

This paper presents a Chainlet based Ear Recognition algorithm using Multi-Banding and Support Vector Machine (CERMB-SVM). The proposed method divides the gray input image into a number of bands based on the intensity of its pixels, resembling a hyperspectral image. It then applies Canny edge detection on each resulting normalized band, extracting edges that represent the ear pattern in each band. The resulting binary edge maps are then flattened, generating a single binary edge map. This edge map is then split into non-overlapping cells and the Freeman chain code for each group of connected edges within each cell is calculated. A histogram of each group of contiguous four cells is calculated, and the results histograms are then normalized and concatenated to form a chainlet for the input image. The resulting chainlet histogram vectors of the images of the dataset are then used for training and testing a pairwise Support Vector Machine (SVM). Experimental results on images of two benchmark ear image datasets show that the proposed CERMB-SVM technique significantly outperforms Principal Component Analysis based methods. In addition, CERMB-SVM also outperforms its anchor chainlet method and state of the art learning based ear recognition techniques.