Abstract:

Gabor filters are widely used in image analysis and computer vision applications. We use a set of multiscale and multiorientation Gabor filters to extract texture features from image blocks of underground sewer duct, and use principal components analysis (PCA) to reduce dimension of the feature vectors. These feature vectors are related to the nature and extend of the damage in the sewerage. Classification of the image blocks is made based on the extracted feature vectors. As the Gabor filters are not orthogonal, suitable parameters should be chosen to minimize redundancy in the extracted image data. The result of k-means clustering verifies the validity of the proposed method. 

Key words: Gabor filter, image recognition, texture feature