Asian Journal of Mathematics and Computer Research

Use of decorative text is extremely preferable to exert a pull on viewer’s attention. The text recognition systems does not always deal with the common text fonts instead also have to work with the various decorated fonts which is used in many fields. The facade of the decorated text is pretty complex compared to the other text forms, and most general recognition systems do not work well with the decorated fonts. In this paper an algorithm has been proposed which work well for the decorated fonts. The proposed algorithm works with neural networks for the recognition and fundamentally consists of three phases. The first phase is preprocessing, second is training of the neural network and the third phase is the testing of the decorated images. Three types of the images have been taken for the testing. First is the web decorated text images that has complex backgrounds, second is web decorated text images with clear background and third is handmade decorated text images. Yet this work has been done with alphabets only. Testing of various multicolor text images has also been done in this. This paper also detects the highest color (among red, blue and green) at a particular point on the text to be recognized.