Color clustering text extraction algorithm for mobile phone images

Variety of approaches for text information extraction from images or video clips have been proposed so far, but none of them is suitable to be implemented over a low computational device, either because of their low accuracy, or slow performance. In this scenario, we propose a Text Extraction algorithm that extracts the text data within natural scene images taken with mobile phone, fast and accurately. The algorithm uses very efficient computations to calculate the Principal Color Components of a quantized image, and separates the main foreground-background colors, after which it extracts the text on the image. We have compared our algorithm with the Otsu algorithm by the use of a commercial OCR, achieving accuracy rates 12% higher, and performing 2 times faster than those algorithms. The proposed approach will be robust against common degradations, such as uneven illumination, or blurring. Therefore, this will be a very attractive algorithm that accurately separates foreground and background from scene text images and works efficiently over low computational resources devices.