Abstract
Background: Generally, the texture information stored in the natural scene images provide some vital clues to the image based on the applications include Content Based Image Retrieval (CBIR), scene understanding, assistive navigation and automatic coding. Still locating the text from complex background and recognizing the characters with different colors are the major issues in image processing applications.
Methods: To overcome these problems, a new structure is proposed in this paper for detecting and recognizing the text strings present in the natural scene images. To execute an efficient framework, the preparation of an input image is necessary that contains two preliminary tasks. It includes preprocessing based on Gaussian filtering and preprocessing based on Histogram equalization. Moreover, an Edge based Intensity Aided Clustering (EIAC) is introduced to detect the text by using the Sobel operator. Then, the neuro-fuzzy algorithm is used classify the text and non-text portions. A new Rule-based Region Map algorithm is introduced for text segmentation, where the necessary features are extracted based on the Local Tetra Patterns (LTrPs). Finally, the characters are recognized with the help of Fuzzy based Relevance Vector Machine (RVMs) classification algorithm. Results: The simulation results obtained by the proposed method are compared with the existing techniques for proving the better performance. The results are analyzed in terms of sensitivity, specificity, accuracy, precision, recall, recognition rate and F-Measure. Conclusion: The proposed framework is fully based on the combination of text detection, classification, segmentation, feature extraction and character recognition. The major advantages of this paper are, accuracy, simplicity and easy to use. Moreover, it provides the best classification results, when compared to the existing techniques.Keywords: Character recognition, Edge based intensity aided clustering (EIAC) algorithm, Local tetra patterns (LTrPs), Relevance vector machine (RVM), Gaussian filter, Histogram equalization, Rule-based region map algorithm, Text detection and segmentation.
Graphical Abstract
11
3