Sharpness and Contrast Based Features for Word-Wise Video Type Classification

Word recognition by a single algorithm from different image types, namely, video-scene, video-caption, natural-scene, mobile camera, Born digital images, etc., is very difficult due to different levels of difficulties. This paper presents a new method combining sharpness and contrast features for classifying different image types at word level using Saturation (S) and Intensity (I) spaces of HSI. For input images, the proposed method performs one dimension filter to smooth each image. It then proposes to perform Maximum Value Difference (MVD) operation to sharpen edge details for the smoothed image. Next, clustering is proposed on enhanced images to identify text candidates. The proposed method extracts sharpness and contrast features in a new way for text candidate images in S and I spaces. K-means clustering is further employed on the extracted set of sharpness and contrast features to obtain different clusters for each space, which results in a feature vector. The feature vector is then fed to an SVM classifier for classification. We use standard datasets, namely, ICDAR 2013, ICDAR 2015 video, natural scene data, caption texts, Born digital data and the images captured by a mobile camera (our own data) to evaluate the performance of the proposed method. Comparative study on classification experiments shows that the proposed method outperforms the existing methods. Recognition experiments before and after classification show that proposed scheme is useful and effective.