REVIEW: ROBUST APPROACH TO DETECT AND LOCALIZE TEXT FROM NATURAL SCENE IMAGES
Khushbu C. Saner, Aditya Deshpande · 2013
In natural scene large amount of information is hidden which can be automatically recognized and processed. To get this automatic segmentation, detection and recognition of visual text entities in natural scene images should be done. For extracting the text from natural scene images, robust approaches like combination of algorithm such as by combining the edge and connected component based are adopted. Efficient outcome is the result of combining. The connected component based algorithm is more robust to scale and lighting condition as compared to edge based. In literature unary component properties and binary contextual component of conditional random field model are preferred to detect nontext components. In region based method, the speed is relatively slow and performance is sensitive to text alignment orientation and for connected component based methods cannot segment text component accurately. Combination of these two models can achieve a far more robust algorithm, which would be invariant to scale, lighting and orientation changes. This robust approach will fail to segment text. The paper emphasizes a strong focus on four methods of detecting and localizing the text already proposed in the literature. A brief review along with comparative analysis and future scope of the algorithm are focused in the paper