Accurate retrieval of textual content from color images has become increasingly important in information retrieval applications. Automated character recognition algorithms, which were mainly designed to process binary document images, frequently fail to perform adequately on color images. Two causes of this are the increased complexity typically found within color images and the lower spatial resolution common to these types of images. Recently, a method of expanding the resolution of text images using a Bimodal-Smoothness-Average (BSA) algorithm was shown to improve character recognition results for grayscale images. In this paper, we extend this grayscale method to enhance color video frame images. The hue-saturation-intensity (HSI) model is used to decouple the color information within an image from its intensity where the BSA algorithm can then be applied. The expanded intensity image is then fused with the hue and saturation to produce a high-resolution image which retains the original color information. This method generates color images that are not only improved visually, but also result in improved character recognition accuracy.
展开▼
