Automatic machine anomaly sound detection is important for machine maintenance in display manufacturing factory. Recently, unsupervised anomaly detection approach based on autoencoder and reconstruction error has been proposed, which has assumption that the network trained only with the normal data produces higher reconstruction errors for the anomalies than the normal inputs. However, most approaches based on autoencoder consist of Convolutional Neural Network (CNN) which deals with the spectrogram of the sound like an image. In this paper, we propose Convolutional Recurrent Neural Network to use the characteristics of the spectrogram. To effectively integrate time and frequency information, we apply a 1‐dimensional convolutional layers and recurrent neural network to the network structure. Also, we introduce a prediction loss to prevent the network from learning to reconstruct anomalies well. Experimental results on the ToyADMOS and MIMII datasets demonstrate that the proposed approach achieves promising performance.

This paper proposes head movement‐based motion blur measurement system for head mounted displays (HMDs). The proposed system simulates the head movement to consider the physical rotation applied to HMDs. Motion blur is quantitatively measured by calculating the moving picture response time (MPRT) from images taken with a high‐speed camera.

The OLED display has been known to have excellent display characteristics such as high response speed, wide viewing angle and color rendering capabilities. However, OLED devices require sustained efforts to improve the OLED lifetime due to a characteristic that the luminance decreases according to elapsed time. In this paper, we proposed hierarchical logo detection and brightness control method to prevent image sticking on OLED TV. We have experimented with this method in FPGA and We prove that the proposed algorithm can increase the lifetime of OLED TV.

In this paper, a photosensor-based latency measurement system for head-mounted displays (HMDs) is proposed. The motion-to-photon latency is the greatest reason for motion sickness and dizziness felt by users when wearing an HMD system. Therefore, a measurement system is required to accurately measure and analyze the latency to reduce these problems. The existing measurement system does not consider the actual physical movement in humans, and its accuracy is also very low. However, the proposed system considers the physical head movement and is highly accurate. Specifically, it consists of a head position model-based rotary platform, pixel luminance change detector, and signal analysis and calculation modules. Using these modules, the proposed system can exactly measure the latency, which is the time difference between the physical movement for a user and the luminance change of an output image. In the experiment using a commercial HMD, the latency was measured to be up to 47.05 ms. In addition, the measured latency increased up to 381.17 ms when increasing the rendering workload in the HMD.

An asynchronous time warp (ATW) is used to reduce a motion-to-photon latency in head mounted displays (HMDs). To implement the ATW, a graphics processing unit (GPU) must allow to perform the time warp during the image rendering. However, typical GPUs do not support the preemption optimized for the ATW. Even though the preemption is performed in the GPUs, visual artifcat like a judder can occur. Therefore, high priced GPUs with high performance optimized for the ATW must be used to enhance the temporal image quality. In this paper, a new field-prgrammable gate array (FPGA) based HMD system is proposed to perform the temporal quality compensation without using ATW. The proposed system can enhance the perceived image quality in the HMD even when using the typical GPUs with low performance. We develop the hardware system on the low priced FPGA handling output images with the 1440X2560 pixel resolution. Experimental results show that the average PSNR of our system is 30.15dB for the temporal image quality and the total execution time of this system is 1.48 ms. Thus, our system can be implmented in real-time with the improved temproal image quality.

디지털 이미지를 외부에서 획득함에 있어 필연적으로 잡음이 발생하게 되며 이는 압축 효율을 떨어뜨리는 요소로 작용한다. 압축 효율을 높이기 위한 방법으로 잡음은 제거하면서 시각에 민감한 엣지는 보존하기 위한 필터 개발은 이미지 처리 분야에서 시대를 불문하고 중요한 이슈였다. C. Tomasi가 제안한 양방향 필터의 경우 연산의 간결함과 필터의 효용성으로 인하여 일반적으로 사용된 필터이나 픽셀 주변 문맥을 고려하지 않고 이미지 내의 모든 픽셀에 사용자가 입력한 변수들이 일괄적으로 적용되어 해당 픽셀에 적합한 변수를 적응적으로 적용하지는 못한다는 문제점이 발생한다. 본 논문은 각 픽셀을 중심으로 소벨 연산을 이용하여 주변의 복잡도를 정의하고 이미지의 화질을 극대화하는 양방향 필터의 변수와의 관계를 함수로 추정하고 이를 통해 변수를 적응적으로 선택하는 알고리즘을 제안한다. 실험에서는 기존의 양방향 필터뿐만 아니라 개선된 성능을 보이는 다른 알고리즘들보다 제안하는 방법이 같은 압축률 대비 PSNR도 뛰어날 뿐만 아니라 엣지 부분에서는 더욱 더 뛰어난 효과가 나타남을 결과를 통해 확인할 수 있다.

This paper presents a novel quantifiable latency measurement system specialized for HMD devices. After analyzing the head position, we implement the rotary platform and the pixel luminance change detector. Then, we accurately measure the latency between the HMD physical movement and an output image change.

Curve subdivision interpolation reconstructs edge well with low complexity, however it lacks of ability to recover texture components, instead. While, neighbor embedding is superior in texture reconstruction. Therefore, in this paper, a novel Super Resolution technique which combines curve subdivision interpolation and neighbor embedding is proposed. First, edge region and non-edge regions are classified. Then, for edge region, the curve subdivision algorithm is used to make two polynomials derived from discrete pixels and adaptive weights are adapted for gradients of 4 different sides to make smooth edge. For non edge region, neighbor-embedding method is used to conserve texture property in original image. Consequently results show that the proposed technique conserves sharp edges and details in texture better, simultaneously.

In this work, we propose a promising approach that compensates OLED driving currents for large-sized OLED displays. With a CCD camera, pixel-to-pixel brightness differences were determined, and applied to compensate OLED driving currents for the purpose of achieving high luminance uniformity. We also implemented the proposed method as a system comprising a FPGA-based timing controller (T-con) and softwares, and the proposed method shows improved luminance uniformity over 20% at the gray level of 32/256.

This paper presents an adaptive subpixel rendering algorithm that uses the luminance difference between the current subpixel and neighboring subpixels. The proposed method reduces both the color fringing artifact and blurring, thereby enhancing greatly the image quality.

A local dimming backlight system achieves both high contrast and low power consumption. The backlight unit (BLU) consists of small blocks. The merits of the local dimming backlight system depend on the size of the blocks. In this work we propose a novel backlight simulator to examine the effect of block_size for edge-type BLUs. The backlight simulator is based on a mathematical model, developed using Matlab 2008b. The development time and cost are reduced by using the proposed backlight simulator.

In various full HD video applications, high-level image enhancement requires localization of overlay text. In this paper, we propose text localization that uses a region-shrinking multistep process to achieve both high accuracy and fast processing in full HD videos.

AbstractHigh contrast ratio and low power consumption for LCDs with LED BLU can be obtained by backlight local dimming with pixel compensation while maintaining image brightness. This technique can greatly increase the contrast ratio of IPS‐mode LCDs. The technique, however, suffers from the side effects called flicker, halo, clipping, and non‐uniformity. The intensity of the side effects is proportional to the amount of the local dimming applied. This paper presents a noble local dimming algorithm which adaptively determines the amount of backlight dimming by characteristics of each block in each scene and feedback values to maximize contrast ratio and power consumption saving effect without noticeable side effects. The experimental results of the proposed algorithm are shown along with those of conventional algorithms.

This paper presents a new backlight dimming approach using dynamic gray‐level error control in liquid crystal displays. The existing methods using the fixed rate of clipped pixels degrade the image quality seriously for some images even after image compensation. The proposed method dynamically chooses the maximum luminance based on image integrity considering image characteristics. Therefore, it preserves the image quality above any desired level, while reducing power consumption as much as possible.

A 2×2 module prototype is developed with four 15‐inch RGBW LCDs. Transmittance of LCD is increased with the low resolution, that is, qXGA(512×RG×384×BW). Using a new RGBW conversion algorithm, white and R/G/B luminance are increased 270% and 130% respectively in comparison with the 15‐inch conventional RGB LCD. An electronics system is implemented and a GUI is developed to support the hardware image control for tiled display.