A new local homogeneity analysis method based on pixel intensities for image defect detection

This paper presents a new technique for defect detection in various images like fabric, paper, metal surface, film surface, nonwoven, coated coil and glass surface etc images. To detect such unstructured defects, it is required to analyze the underlying texture information of the image. All of these images have some similar texture properties. These texture features can be achieved through local homogeneity analysis (H-image) such as edges and boundaries which arise due to sudden changes in intensity levels of pixels. The defected regions can be one or more in a texture based image which may vary in shape and gray levels as well, thus the defect detection is a challenging task. Traditional homogeneity based approaches gives the fine edge details as well, some of them can be treated as a defect which is not a defect. So, this paper proposes a new technique to find the local homogeneity image using pixel intensity differences among neighboring pixels. Thereafter for thresholding a DWT (Discrete Wavelet Transform) and Hotelling T² model are used. The performance of the proposed approach has been evaluated against various measures like precision, specificity, recall, accuracy and error rate. In comparison to the existing H-Image the proposed approach had 98% accuracy when tested on 100 different types of images.