Chung Yen Su

Image quality improvement for electrophoretic displays by combining contrast enhancement and halftoning techniques

The electrophoretic display (EPD) has dominated the panel used in electronic readers. However, getting more than 16 observable gray levels on an EPD is still a big challenge. A true color image should be converted into a 16 gray level image for displaying it on an EPD. In this paper, we propose an image processing tool that combines contrast enhancement and the modified Nasik pattern halftoning to deal with the problems in the image conversion chain. The experimental results show that the processed image has preserved more image textures after the image data are quantized into 16 gray levels1.

Highly effective iterative demosaicing using weighted-edge and color-difference interpolations

Demosaicing is a process of obtaining a full-color image by interpolating the missing colors of an image captured from a single sensor color filter array. This paper provides an effective and low-complexity iterative demosaicing algorithm applying a weighted-edge interpolation to handle green pixels followed by a series of color-difference interpolation to update red, blue, and green pixels. Based on our experiments of images, we enable the algorithm a well-designed stopping condition and pre-determine the proper weights of interpolation. Experimental results show that the proposed method performs much better than three state-of-the-art demosaicing techniques in terms of both computational cost and image quality. In comparison to the algorithm of successive approximation, the algorithm proposed here reduces mean squared error up to 14.5% while requiring computational cost only 22% on average. That is, it takes less time but performs better.

An enhanced segmentation on vision-based shadow removal for vehicle detection

We present an enhanced segmentation method to reduce the interference of shadows for vehicle detection. The main advantage of the proposed method is its low-complexity. We use only the luminance of an image for shadow removal and keep the chrominance components of the image intact. The luminance of the current image is enhanced and each pixel is compared with a pixel-dependent threshold for locating shadow regions. With that, the shadow regions can be located more accurately and the moving objects can be extracted more completely. Experimental results verify the proposed approach and show that it is helpful for vehicle detection.

Effective demosaicing using subband correlation

An effective demosaicing is presented to restore the missing pixels of images captured from single-sensor cameras. The proposed method uses sub-image subband correlation to enable a good initial interpolation and accurate edge detection. Experimental results show that the proposed method not only has less computation but also produces higher quality than three state-of-the-art iterative demosaicing methods.

An enhanced detection algorithm for all-zero blocks in H.264 video coding

A block is called as an all-zero block (AZB) if all its transformation coefficients are quantized to be zero. Provided that an AZB can be detected early, the processes of transformation and quantization on an AZB can be omitted. This leads to significant redundant computations being skipped and thus speeds up the coding of a video sequence. In this paper, a more precise threshold value than previous methods is proposed to increase the number of AZBs detected. The threshold value is cautiously derived from relative theories and no assumption is adopted, ensuring that video quality is not degraded. A comparison to different methods on detectable ranges is graphically illustrated to show the improvement of the proposed method. The computational complexity of the proposed algorithm is analyzed. Experimental results show that the proposed algorithm outperforms the previous methods in all cases and achieves major improvement of computation reduction in the range from 4% to 71.3% compared to previous methods. The larger the quantization is, the larger the computation reduction is.

A Kinect-based handwritten digit recognition for TV remote controller

Human-machine interaction is a popular research filed recently. Microsoft Kinect can provide us an economical way for tracking the skeleton of the human body. In this study, we present an efficient Kinect-based method for a potential application of TV remote controller. The proposed method consists of two parts: four direction (up/down/left/right) recognition and handwritten digit recognition. The developed recognitions include five main steps: initialization, tracking skeleton, judging the start condition, recording the path, and recognition. We use an intuitive way to start the event and achieve high accurate recognition for the users. We apply the least square method for the four direction recognition and a support vector machine library for the handwritten digit recognition. Experimental results show that the proposed system has great potential on future human-machine development.

Sub-pixel Edge Detection of LED Probes Based on Canny Edge Detection and Iterative Curve Fitting

In recent years, the demands of LED are increasing. In order to test the quality of LEDs, we need LED probes to detect it, so the accuracy and manufacturing methods are attracted more attention by companies. LED probes are ground by people so far. When processing, we often consider the angle and radius of a probe (the radius is between 0.015 mm and 0.03 mm), so it is hard to balance between precision and quality. In this study, we proposed an effective method to measure the angle and radius of a probe. The method is based on Canny edge detection and a curve fitting with iteration. Experimental results show the effectiveness of the proposed method.

An Effective and Fast Lane Detection Algorithm

Lane detection is crucial for autonomous driving. In this paper, we present an effective and fast lane detection algorithm. The proposed algorithm includes three novelties. First, we set a region of interest (ROI) appropriate to reduce nonessential cost of computation. Second, we determine a real midpoint between two road lines for each frame. The midpoint can be used to classify the candidates of lane marking points to right and left effectively. Finally, we use a temporal trajectory strategy to avoid the failure of lane detection, which is generally caused by shadows of bridges or neighboring vehicles. Experimental results show that the proposed algorithm can label the location of lane marking accurately and fast. It processes a frame only 16 ms and can solve the problems caused by lighting change, shadows, and vehicle occlusions.

An FPGA implementation of chaotic and edge enhanced error diffusion

Digital halftoning plays a central role in getting more observable grey-levels for either the innovative electronic paper or other less level devices. The hardware implementation of digital halftoning is, however, seldom fully explored. In this paper, we propose a novel implementation of digital halftoning by means of error diffusion. The proposed scheme not only can perform a new method called chaotic and edge enhanced error diffusion, but also can be reduced to perform the conventional Floyd-Steinberg error diffusion. Best of all, our new scheme can produce halftone images with lower worm-like artifacts and sharper image edges. This scheme is mainly composed of four components: gradient-based edge detection, chaotic threshold generation, edge enhanced quantization, and error diffusion. Each circuit design of the four components is illustrated for the first time. Besides, we demonstrate the hardware performance of our scheme by using a field programmable gate array (FPGA) chip to offer possibly further applications.

Adaptive Early Termination for Fast H.264 Video Coding

The H.264 standard applies several powerful coding methods to obtain high compression efficiency. However, it requires a lot of computation especially in variable block-size motion estimation. To reduce the motion estimation redundancy more effectively, an adaptive early termination algorithm is proposed in this paper. The proposed algorithm dynamically changes the thresholds for different coding modes according to video content. With the proposed method, many zero motion blocks can be predicted, the corresponding motion estimation can stop early, and the remaining computation can be omitted. Simulation results show that the proposed method can averagely reduce the entire coding time up to 14.38% and the motion estimation time up to 21.82% at the price of negligible coding loss.