Text Recognition By Using Character Descriptor And SVM Classifier

Generally, the images captured by Camera has many different shapes, sizes, colours, text, non-text etc regions which very complex the Camera-based scene images usually have background which is very complex. The existing system is very sensitive to font scale changes and background interference with low accuracy. The most important aim of this system is based on character recognition method. Separating text or characters from captured scene images or videos is a very difficult task because of different text styles, fonts, patterns and variable background image interferences. We are proposing in this paper that a process of natural scene text recognition from selected text regions from a natural image. In text detection, we detect text from any natural image by using MSER (Maximally Stable Extremal Region) algorithm. MSER contains the text region from an image; for text recognition and the proposed system uses character descriptor which is very effective in extracting image. The local features descriptor HOG is suitable and compatible with all main points’ detectors from interested region. Our method of text recognition from detected text regions is very compatible with an application of mobile devices. The demo system which developed is completely based on Android operating system. The Proposed system exactly extracts text from any natural scene image with background interference. The demo system gives us details of algorithm design and performance improvements of scene text extraction.