ePrints@IIScePrints@IISc Home | About | Browse | Latest Additions | Advanced Search | Contact | Help

Color Components Clustering and Power Law Transform for the Effective Implementation of Character Recognition in Color Images

Giritharan, Ravichandran and Ramakrishnan, AG (2017) Color Components Clustering and Power Law Transform for the Effective Implementation of Character Recognition in Color Images. In: PROGRESS IN INTELLIGENT COMPUTING TECHNIQUES: THEORY, PRACTICE, AND APPLICATIONS, VOL 1, 2018, INDIA, pp. 131-139.

Full text not available from this repository.
Official URL: https://doi.org/10.1007/978-981-10-3373-5_12

Abstract

The problem related to the recognition of words in the color images is discussed here. The challenges with the color images are the distortions due to the low illumination, that cause problems while binarizing the color images. A novel method of Clustering of Color components and Power Law Transform for binarizing the camera captured color images containing texts. It is reported that the binarization by the proposed technique is improved and hence the rate of recognition in the camera captured color images gets increased. The similar color components are clustered together, and hence, Canny edge detection technique is applied to every image. The union operation is performed on images of different color components. In order to differentiate the text from background, a separate rectangular box is applied to every edge and each edge is applied with Power Law Transform. The optimum value for gamma is fixed as 1.5, and the operations are performed. The experiment is exhaustively performed on the datasets, and it is reported that the proposed algorithm performs well. The best performing algorithm in ICDAR Robust Reading Challenge is 62.5% by TH-OCR after pre-processing and 64.3% by ABBYY Fine Reader after pre-processing. The proposed algorithm reports the recognition rate of 79.2% using ABBYY Fine Reader. This proposed method can also be applied for the recognition of Born Digital Images, for which the recognition rate is 72.5%, earlier in the literature which is reported as 63.4% using ABBYY Fine Reader.

Item Type: Conference Paper
Series.: Advances in Intelligent Systems and Computing
Publisher: SPRINGER-VERLAG BERLIN
Additional Information: Copyright for this article belongs to Springer
Keywords: Color component clustering; Canny edge detection; Power law transform; Binarization; Word recognition
Department/Centre: Division of Electrical Sciences > Electrical Engineering
Date Deposited: 12 Jan 2019 10:50
Last Modified: 12 Jan 2019 10:50
URI: http://eprints.iisc.ac.in/id/eprint/61263

Actions (login required)

View Item View Item