Comparative Evaluation of Handcrafted Image Descriptors with IBK Classifier for Automated Skin Disease Classification

The right and timely diagnosis of dermatological problems continues to be critical in order to come up with requisite medical therapies alongside the treatment regimes. They evaluate IBK (Instance-Based k -Nearest Neighbor) as a classification system on analysis of six feature extraction method that include Auto Color Correlogram Filter (ACCF), Binary Patterns Pyramid Filter (BPPF), Fuzzy Color and Texture Histogram Filter (FCTHF), Fuzzy Opponent Histogram Filter (FOHF), Pyramid Histogram of Oriented Gradients (PHOG) and Simple Color Histogram Filter (SCHF) in recognizing skin disease patterns. Users used the Skin Diseases Image Dataset of Kaggle with 38,000 high-resolution images having different skin diseases classifications. Testing performance metrics were composed of Accuracy, Precision, Recall, ROC along with PRC and Execution Time, which were considered as standard. BPPF+IBK and SCHF+IBK produced the best classification results of 92.98% and FCTHF+IBK had the best recall of 0.96 and execution time of 0.01 seconds. The items of information acquired by FOHF+IBK detection revealed the outstanding PRC performance of 0.96. The study has shown some significant results regarding the relationship between accuracy indices and computational cost that can be used to develop novel lightweight systems of skin diseases detection to support clinics.