A Comparative Study of Feature Extraction Methods in Image Classification Using Convolution Neural Network Model
Divya Jennifer Dsouza, Anisha P Rodrigues · 2023
Image classification is the supervised classification in which one assigns labels to images based on its characteristics or features. The term Outlier which is also referred as anomaly are points, patterns or sequences of data that do not fall into the normal behaviour of the system under consideration. Outlier detection refers to the study of identifying the genuine outlier within a given data set. This paper aims to present a deep learning convolution neural network which is based on outlier detection technique with applied gabor, LBP and SIFT filters for feature extraction that can detect outliers by assigning weights to selected features such as to prioritise its significance. A comparison is made with the traditional CNN and applied filters on two image data-sets. The first dataset is the publicly available malaria parasites blood smear sample dataset and the the second dataset is the publicly available facial expressions dataset. Classification related evaluation metrics were applied to the results for analysis. Our experimental performance indicate that the applied filters combined with CNN performs considerable better in detecting existing anomalies with good accuracy but the training time to create features exceeds the traditional approach.