Text Extraction from Images

Automatic image annotation, structuring of images, content-based information indexing and retrieval are based on the textual data present in those images. Text extraction from images is an extremely difficult and challenging job due to the variations in the text such as text scripts, style, font, size, color, alignment and orientation; and due to extrinsic factors such as low image contrast (textual) and complex background. However, this is realizable with the integration of the proposed algorithms for each phase of text extraction from images using java libraries and classes. Initially, the pre-processing phase involves gray scaling of the image, removal of noise such as superimposed lines, discontinuities and dots present in the image. Thereafter, the segmentation phase involves the localization of the text in the image and segmentation of each character from the entire word. Lastly, using the neural network pattern matching technique, recognition of the processed and segmented characters is done. Experimental results for a set of static images confirm that the proposed method is effective and robust. Keywords— Image Pre-processing, Binarization, Localization, Character Segmentation, Neural Networks, Character Recognition.