Data Analytics and Learning: Proceedings of Dal 2018 (Lecture Notes in Networks and Systems #43)

By: P. Nagabhushan and D. S. Guru and B. H. Shekar and Y. H. Sharath Kumar

Sign Up Now! Already a Member? Log In
You must be logged into Bookshare to access this title. Learn about membership options, or view our freely available titles.

Synopsis: This paper describes a method to localize and recognize seven-segment displays on digital energy meters. Color edge detection is first performed on a camera-captured image of the device which is then followed by a run-length technique to detect horizontal and vertical lines. The region of interest circumscribing the LCD panel is determined based on the attributes of intersecting horizontal and vertical lines. The extracted display region is preprocessed using the morphological black-hat operation to enhance the text strokes. Adaptive thresholding is then performed and the digits are segmented based on stroke features. Finally, the segmented digits are recognized using a support vector machine classifier trained on a set of syntactic rules defined for the seven-segment font. The proposed method can handle images exhibiting uneven illumination, the presence of shadows, poor contrast, and blur, and yields a recognition accuracy of 97% on a dataset of 175 images of digital energy meters captured using a mobile camera.