Text Detection in Multi-Oriented Natural Scene Images

---------------------------------------------------------------------***--------------------------------------------------------------------Abstract With the growing number of digital multimedia libraries, the need to efficiently index, browse and retrieve multimedia information is increased. Text embedded in images and video frames can help to identify the image information (e.g. somebody's name appearing in an image) or to display information which is independent of the image (e.g. important news during the transmission of a movie). In common, text in images can be categorized into two groups: artificial text and scene text. Scene text is part of the image, and appears unintentionally, like in traffic signs etc. whereas artificial text is created separately from the image and is laid over it in a later stage, like the name of a journalist at some point in a news program. Artificial text is usually a very good key to index image or video databases. This paper aims to detect text in scene images of multi-orientations, i.e., the images containing text in multiple text lines and the images with problems of skewed angle and distorted images. The preprocessed algorithm uses skew correction and image enhancement algorithms to preprocess the images for better recognition accuracy. Every image has to be pre processed before further processes. The pre processing technique is also tested for the accuracy and time taken to perform the detection.