Image processing algorithms for improved character recognition and components inspection

The importance of inspection process has been magnified by the requirements of the modern manufacturing environment. A variety of approaches for automatic inspection of machine parts have been reported over last two decades. In this work it is targeted to develop an automated machine vision software system, which is used to inspect automotive parts like PE pumps after assembly of different components. Inspection is carried out to detect missing components, misalignment of components and out of tolerance of components. This work proposes some methodologies for Optical Character Recognition (OCR) and feature extraction. An improved Local Mean-Gradient thresholding algorithm is proposed and implanted to recognize the characters like type number and serial number printed on the automotive parts like PE-PUMP, irrespective of the variation in background colors, industrial noises like dust particles, oily surfaces etc. The proposed methodologies have been implemented and explained here with some generic examples which overcome various industrial constraints. In this work, it is proved that automatic selection of threshold based on the background colors has improved the OCR performance to 100% recognition. The second part focuses on the extraction of suitable features in order to obtain a good pattern matching result even under various industrial constraints. Laplacian method of edge which is based on difference in gray level is more sensitive to noise. The second part gives an improved edge detection method using Laplacian operator.