Wang-Jun Kyung

Real-time multi-scale Retinex to enhance night scene of vehicular camera

Recently, image processing has been commonly applied to various parts of the vehicle industry. In particular, the use of vehicle cameras gives the driver clear view for occluded regions to the front or rear of the vehicle. However, images from the camera have low visibility at night due to the lack of ambient illumination. To enhance the visibility of images, generic methods using global functions, such as the gamma function, can be used for real time processing, but these methods cannot improve local visibility. In this paper, we propose real time processing of night scene images based on modified multi-scale Retinex(MSR) to enhance the visibility of images from vehicle camera. This method uses integer operations and bit-sifting for real-time computation. In addition, we apply simple blur and mean filters instead of Gaussian convolution to reduce computational cost. Color balance is preserved by obtaining the surround image for the luminance channel only. In the experiments, processing speed approximates to 30 frame per second(FPS) and dark regions in the frames are improved through enhancement of local visibility.

Estimation of reflectance based on properties of selective spectrum with adaptive Wiener estimation

To accurately represent the colors in a real scene, a multi-channel camera system is necessary. One of the applications of the data acquired with the multi-channel camera system is the spectral reflectance estimation. One of the most widely used methods to estimate the spectral reflectance is the Wiener estimation. While simple and accurate in controlled conditions, the Wiener estimation does not perform as well with real scene data. Therefore, the adaptive Wiener estimation has been proposed to improve the performance of the Wiener estimation. The adaptive Wiener estimation uses a similar training set that was adaptively constructed from the standard training set according to the camera responses. In this paper, a new way of constructing such similar training set using the correlation between each spectral reflectance in the standard training set and the first approximation of the spectral reflectance that was obtained by the Wiener estimation is proposed. The experimental results showed that the proposed method is more accurate than the conventional Wiener estimations.

Improved color reproduction based on CIELAB color space in integrated multi-scale retinex

Recently, tone reproduction is widely used in the field of image enhancement and HDR imaging. This method is especially used to provide the proper luminance so that captured images give the same sensation as the scene. As a result, we can get high contrast and naturalness of colors. There is ample literature on the topic of tone reproduction that has the objective of reproducing natural looking color in digital images. In recent papers, IMSR (Integrated multi-scale Retinex) shows great naturalness in the result images. Most methods, including IMSR, work in RGB or quasi-RGB color spaces, although some method adopted the use of luminance. This raises hue distortion from the point of the human visual system, that is, hue distortion in CIELAB color space. Accordingly, this paper proposes an enhanced IMSR method in a device-independent color space, CIELAB, to preserve hue and obtain high contrast and naturalness. In order to achieve the devised objectives, a captured sRGB image is transformed to the CIELAB color space. IMSR is then applied to only L* values, thus the balance of colors components are preserved. This process causes unnatural saturation, therefore saturation adjustment is performed by applying the ratio of chroma variation at the sRGB gamut boundary according to the corrected luminance. Finally, the adjusted CIELAB values are transformed to sRGB using the inverse transform function. In the result images of the proposed method, containing both high and low luminance regions, visibility in dark shadow and bright regions was improved and color distortion was reduced.

Color enhancement for faded images based on multi-scale gray world algorithm

This paper presents a color enhancement algorithm based on multi-scale gray world algorithm for faded images. First, the proposed method adopts local process by using multi-scale mask. The coefficients for each multi-scale mask are obtained to apply the gray world algorithm. Then, integrating the coefficients with weights is performed to calculate correction ratio for red and blue channels in the gray world assumption. Finally, the corrected image is obtained by applying the integrated coefficients to the gray world algorithm. In the experimental results, the proposed method reproduces better colors for both wholly and partially faded images compared with the previous methods.

Selective skin tone reproduction using preferred skin colors

People in a color image, particularly facial patterns are important and interesting visual object. Thus, color reproduction for skin tone becomes the foremost treatment, because skin color is a key memory color in color application systems. Previous studies suggested skin color reproduction by mapping only to the center value of preferred skin region. However, it is not suitable to determine one preference color because preference color from the observers preference test is not dominant. In this paper, selective skin color reproduction using preferred skin colors is proposed. The proposed method first defines preferred skin colors according to their luminance level. Next skin color is detected by using skin and feature map. After that, each corresponding preferred skin color is determined for each skin region. Finally, input skin color is proportionally mapped toward preferred skin color according to the difference between input skin color and preferred skin color, smoothly reproducing skin color. Through the experimental results, the proposed method achieves preferred skin color reproduction.

Spatial gamut mapping for preserving the details of an image

The spatial gamut-mapping algorithm (SGMA) overcomes the drawbacks of the widely used color-by-color methods. Spatial gamut mapping can preserve detailed information in original images by performing adaptive gamut mapping in surrounding pixels within the image. However, spatial gamut mapping can result in hue shift and the halo effect. In addition, it only preserves the boundary information outside the color gamut; the resulting gamut-mapped image does not sufficiently preserve the detailed information in the input image. In this paper, we propose an SGMA that utilizes details of the input image. Our approach improves detail that is not effectively represented with conventional spatial gamut mapping. This is done by taking an original image and first implementing gamut mapping of the input image. Then, the details of the input image and gamut-mapped image are extracted. By examining the out-of-gamut region, the details of the input image can be preserved when these values are added to the gamut-mapped image. The resulting image is obtained by clipping out-of-gamut pixels, since these pixels are generated in the process of preserving details. We demonstrated that images obtained using the proposed method are more similar to the input images, compared to images obtained using conventional methods.

Simultaneous color matching in stereoscopic images based on image decomposition

Color discrepancies between the left and right images from a stereoscopic camera cause many problems, including a reduction in the three-dimensional effect and increased visual fatigue. Thus, color matching of stereoscopic images is very important for three-dimensional display systems. Therefore, this paper proposes simultaneous color matching based on image decomposition. The stereoscopic images are first decomposed into various levels, and the colors are then matched using cumulative histogram matching and a Retinex algorithm. Experimental results show that the color matching performance of the proposed method is better than that of previous methods.

Moire reduction using inflection point in frequency domain

Digital still cameras generally use an optical low-pass filter(OLPF) to enhance the image quality by removing high spatial frequencies causing aliasing. While eliminating the OLPF can save manufacturing costs, images captured without using an OLPF include moiré in the high spatial frequency region of the image. Therefore, to reduce the presence of moiré in a captured image, this paper presents a moiré reduction method without the use of an OLPF. First, the spatial frequency response(SFR) of the camera is analyzed and moiré regions detected using patterns related to the SFR of the camera. Using these detected regions, the moiré components represented by the inflection point between the high frequency and DC components in the frequency domain are selected and then removed. Experimental results confirm that the proposed method can achieve moiré reduction while preserving detail information.

Moire Reduction in Digital Still Camera by Using Inflection Point in Frequency Domain

Digital still camera generally uses optical low-pass filter(OLPF) to enhance its image quality because it removes high spatial frequencies causing aliasing. However, the use of OLPF causes some loss of detail. On the other hand, when image are captured by using no OLPF, the moir is generally existed in high spatial frequency region of an image. Therefore, in this paper, moir reduction method in case of using no OLPF is suggested. To detect the moir, spatial frequency response(SFR) of camera was firstly analyzed by using ISO 12233 resolution chart. Then, moir region is detected by using the patterns that are related to the SFR of camera. next, this region is analysed in the frequency domain. Then, the moir is reduced by removing its frequency component, which represents inflection point between high frequency and DC components. Through the experimental results, it is shown that the proposed method can achieve moir reduction with preserving the detail.

Accurate Wiener Estimation by Constructing a Similar Training Set Based on Spectral Correlation

An image is generally acquired from traditional digital camera with three channels, such as red, green, and blue. However, an RGB image cannot fully represent real scene. To accurately represent the colors in a real scene, a multi-channel camera system is used to estimate spectral reflectance. Wiener estimation is widely used methods for estimating the spectral reflectance. While simple and accurate in controlled conditions, the Wiener estimation does not perform as well with real scene data. Therefore, the adaptive Wiener estimation has been proposed to improve the performance of the Wiener estimation. It uses a similar training set that was adaptively constructed from the standard training set. In this paper, a new way of constructing similar training set is proposed by using the correlation between spectral reflectance in the standard training set and the approximated spectral reflectance by the Wiener estimation. The experimental results show that the proposed method is more accurate than the conventional Wiener estimations.