Gradient Feature-Based Classification of Patterned Images

Abstract

Image classification is the task of assigning a class to an image. It has a wide range of applications: image and video retrieval, object tracking, object recognition, Web content analysis, number plate recognition, OCR in banking systems, etc. Color, texture, gradient, shape, keypoint descriptors, etc. are the various features being used for the image classification. A patterned image is an image in which selected pattern is repeated, for example, horizontal stripes, vertical stripes, polka dots, geometric shapes, etc. Gradient feature plays a vital role in distinguishing the different patterns. Therefore, in the proposed approach, gradient features are used for the classification of patterned images like cloth patterns (vertical stripes, horizontal stripes, polka dots, etc.), English characters (capital and small alphabets) and numerals (0–9) and geometric shapes (square, triangle, etc.). The different patterns recognized in the present paper show the versatility of the approach. It can be applied to many of the real-time applications like number plate recognition, cloth pattern recognition and retrieval. The proposed approach achieves the accuracies of 95.4, 93.5, 91.4 and 92% on standard datasets describable texture dataset (vertical stripes, polka dots), EnglishImg dataset (small and capital English alphabets), numerals dataset (0–9) and geometric shapes (triangle, square) dataset, respectively.