High-definition Image Processing: Dynamic Range Expansion and Text Localization

Abstract

This thesis deals with high-definition image processing: dynamic range expansion and text localization. To expand dynamic range, local dimming for liquid crystal displays (LCDs) and high dynamic range (HDR) imaging techniques are presented. Text localization is an important information extraction method in high-definition image. LCDs have been increasingly used in displays due to their low cost, lack of radiation emission, high video quality playback and light weight. The local dimming backlight technique enables LCDs to present HDR images with high contrast ratio (CR) and low power consumption by dividing the backlight into several local blocks and modulating them individually according to the input image. However, high CR with local dimming causes clipping artifacts. Thus we propose a novel local dimming algorithm to expand dynamic range with reduced clipping artifacts. Generally, a conventional 8-bit image can be processed to an HDR image by inverse tone mapping operators. Among the operators, brightness discrimination mapping by applying brightness adaptation model has provided a good fit to electro-physiological and psychophysical measurements. We use the brightness adaptation model to derive an optimum brightness discrimination mapping algorithm for HDR images. The proposed algorithm maximizes local and global brightness discrimination range by exploiting characteristics of the human visual system. Enhancement of details is verified by visualizing HDR images from dark to bright regions. Improvement of dynamic range is quantified by measuring increased discrimination ratio. Overlay text provides helpful clues for interpreting the video content in various kinds of spatially high-definition video applications. Thus detection and localization of overlay text are important. This thesis presents an overlay text detection and localization algorithm that satisfies the requirements for both accuracy and processing speed for high-definition video. In overlay text detection, we obtain a 1st order derivative image from the intensity and solve an error minimization problem to determine the threshold of text detection. Then overlay text localization is performed through stepwise block-based processing, which improves the reliability of the determination of the overlay text region. The proposed algorithm is suitable for detecting and localizing overlay text in high-definition video because its computational complexity is efficiently minimized by reducing iterative operations. Experimental results show that our algorithm has high accuracy in both still images and videos with various kinds of overlay text and background.