Text Extraction from Natural Scene using PCA

Many techniques and algorithm have been developed to solve the problem of text extracted from natural scenes. Text extraction is emerging and challenging era in the computer vision. Text which is embedded into the image contains semantic information which is used in many other applications such as information retrieval of complex images, robot navigation, useful for visually impaired persons, street signs, automatic read the sign board and use in so many other applications. Most of the research work in this area has been done only on printed text, a very few research is addressing the LED scene text. Scene text is difficult to extract due to blur image, variations in color, noise problem, complex background, discontinuity, poor lighting conditions, and variation in illumination. LED is Light Emitting Diode which is widely used in displaying the information in LED boards. Now days LED display that is natural scene is being widely used for displaying announcements, sign boards, banners for displaying information. To extract the text from the LED display is not an easy task, it is very complex due to its discontinuity. A matrix of segments is used to display the character of LED, which is combined together to generate an LED text. So, The aim of this paper to propose a technique to extract an LED text from natural scene image. In the preprocessing step, The RGB input image will be converted to a grayscale image, image is binarization and noise is removed. Then FFT and FFT shift is used to extract the text region because the text is generally found in higher frequency and it is the fastest method. The text or non-text region is classified. Finally, apply the template matching method is developed using PCA, which is used to recognize the extracted text and display in boundary boxes. The experimental results of the proposed method show the extraction rate is 73.25. Keywords— Connected Component method, FFT method, FFT shift, MSER, Morphological operations, PCA, Text extraction.