Text Detection using Multilayer Separation in Real Scene Images

Robust extraction of text from scene images is essential for successful scene text recognition. Scene images usually have non-uniform illumination, complex background and existence of text-like objects. Due to non-uniform illumination the contrast of text pixels sinks and merges with background. As a result the common assumption of a homogeneous text region on a nearly uniform background cannot be maintained in real application. We proposed a text extraction method that utilizes the available contrast of text pixels. Initially the colours of input images were reduced to small numbers using quantization method. Next, colour reduced images were converted to gray scale and pixels intensity values were estimated. Based on the intensity values, the pixels were separated into three layers of connected components. Missing texts were recovered using image binarization technique. Then the geometrical features of the entire connected components in each layer were analyzed to extract the text regions from nontext regions. And finally layers were merged together and using a Sobel edge projection profile the text line was verified. To show the effectiveness of our proposed method scene text dataset with different font shapes and sizes, direction, background and contrast were tested and experiment results showed high performance and detection rate.