Feature Extraction of Unprinted Fabric for Defect Detection Using Gabor Filter, HOG and SVM

This paper presents a precise and efficient method for feature extraction of unprinted fabric for defect detection using a combination of Gabor filters, Histogram of Oriented Gradients (HOG), and Support Vector Machine (SVM). The proposed approach utilizes Gabor filters with various scales and orientations to capture local textural features of the fabric at different granularities, enhancing the discriminative power for defect identification. Subsequently, HOG descriptor is employed to encode the distribution of oriented gradients, providing robust features against illumination variations. Finally, an SVM classifier is trained to effectively distinguish between defect and non-defect regions based on the extracted features. The proposed method is extensively evaluated on a dataset of unprinted fabrics with diverse defect types. The experimental results demonstrate the superiority of the proposed approach in achieving superior accuracy for defect detection in unprinted fabrics compared to existing methods, highlighting its potential for real-world applications in the textile industry.