1：Experimental Design and Method Selection:

The methodology involves preprocessing ear images using a median filter and normalization, applying a bank of binarised statistical image features (B-BSIF) filters at multiple scales to derive response images, computing and concatenating histograms of these responses, projecting the histograms into a linear discriminant analysis (LDA) subspace using whitened linear discriminant analysis (WLDA) for dimensionality reduction and increased discriminability, and classifying the resulting feature vectors using a K-NN classifier with chi-square or cosine distance metrics.

2：Sample Selection and Data Sources:

Three public ear databases are used: IIT Delhi-1 (493 images, 125 subjects), IIT Delhi-2 (793 images, 221 subjects), and USTB-1 (185 images, 60 subjects). Images are preprocessed to standardized sizes (e.g., 180x50 or 150x80 pixels).

3：List of Experimental Equipment and Materials:

No specific hardware is mentioned; the experiments are computational, using learned BSIF filters and standard image processing techniques.

4：Experimental Procedures and Operational Workflow:

For each database, images are preprocessed (median filter and normalization), convolved with B-BSIF filters at various scales, histograms are computed and normalized, concatenated, projected using WLDA, and classified with K-NN. Multiple experiments vary filter parameters (window size l and bit string n) and compare different representations (MS-BSIF and DMS-BSIF).

5：Data Analysis Methods:

Recognition rates are calculated as the average accuracy over permutations, using chi-square distance for MS-BSIF and cosine distance for DMS-BSIF in the K-NN classifier.