Recent advancements in computer vision based techniques have grown drastically which are adopted widely in real-time applications such as image retrieval, automatic street sign analysis and license plate detection &recognition. In such type of applications, text detection and recognition is considered as most promising and challenging task for several real-time applications. Natural scene image text has complex backgrounds, image blurriness, occlusions, font-style variations, illumination variation etc. resulting in increased complexity in text recognition from natural scene images. Considerable amount of research work has been carried out in this research field but obtaining desired and satisfying accuracy for text detection and extraction with reduced error is still remains a challenging task for researchers. In this work, we present an adaptive approach for text recognition and extraction from natural scene images. In order to perform text detection, scale varied Ada Boost classifier is used. Furthermore, text-extraction process is discussed where Maximally Stable Extremal Regions (MSER) and LBP (Local binary pattern) are combined to obtain the significant feature vector. Later, k-means clustering and SVM (Support Vector machine) are used for splitting text region and finally, deep convolutional neural network is used for classification of detected text.
Volume 11 | 04-Special Issue
Pages: 836-846