Kyungman Kim

Bio: Kyungman Kim is an academic researcher from Yonsei University. The author has contributed to research in topics: Tone mapping & Stairstep interpolation. The author has an hindex of 3, co-authored 3 publications receiving 69 citations.

Natural hdr image tone mapping based on retinex

Abstract: In this paper, new k factor decision method and highlight compression operator are proposed to enhance the appearance and naturalness of rendered High Dynamic Range (HDR) images. The retinex algorithm is one of the outstanding local operators, which well preserves local contrast in highlights. However, the retinex algorithm gives a worse overall appearance and undistinguishable dark area contrast than global operators or other local operators in some cases. The most prominent improvement of the proposed method is that the decision method of the k factor, which is one of the parameters in retinex algorithm, is proposed by using the dynamic range in images. The proposed parameter decision method enhances the overall quality and preference of the image and solves any parameter setting problems. Also, dark area details become more distinguishable by the highlight compression operator. According to the results of many HDR image experiments, the proposed method shows better rendering in terms of naturalness and dark area details than the previous tone-mapping algorithm.1.

A new edge directed interpolation algorithm using accurate estimation of edge directional covariance

Abstract: This paper proposes an edge-directed interpolation algorithm to enhance the quality of natural images which are captured by low-resolution camera installed on car or CCTV. Based on the accurate estimation of edge directional covariance between low-resolution and high-resolution image, local covariance coefficients extracted from the low-resolution image has been adapted for the interpolation to obtain the high-resolution image. DCT (Discrete Cosine Transform) kernel function is used in order to reflect the multi-directional edge accurately without increasing of complexity. Simulation result shows that our new interpolation algorithm significantly improves the subjective quality of the interpolated images compared with conventional linear interpolation and NEDI one. It also demonstrates the improvements of objective metrics such as PSNR, SSIM(structural similarity index measurement) and WEA (Wiener filter coefficients Estimation Accuracy) which are used for the accuracy estimation of directionality.

Adaptive tone-mapping operator for HDR images based on image statistics

Abstract: In this paper, we present a novel segmentation method for displaying high dynamic range image based on K-means clustering. The new segmentation method uses statistical features of an image in a logarithmic luminance domain. Each divided region is applied to different global tone mapping operators respectively. The global tone mapping operator is a logarithmic tone mapping with a different user parameters. The parameters for applying to each region are calculated using a centroid which is obtained from K-means clustering. According to results of many HDR image experiments, we demonstrate that our method is faster than other local tone mapping operators and improves an image rendering performance in terms of dark area details and contrast enhancement.

Fast multi-exposure image fusion with median filter and recursive filter

Abstract: This paper proposes a weighted sum based multi-exposure image fusion method which consists of two main steps: three image features composed of local contrast, brightness and color dissimilarity are first measured to estimate the weight maps refined by recursive filtering. Then, the fused image is constructed by weighted sum of source images. The main advantage of the proposed method lies in a recursive filter based weight map refinement step which is able to obtain accurate weight maps for image fusion. Another advantage is that a novel histogram equalization and median filter based motion detection method is proposed for fusing multi-exposure images in dynamic scenes which contain motion objects. Furthermore, the proposed method is quite fast and thus can be directly used for most consumer cameras. Experimental results demonstrate the superiority of the proposed method in terms of subjective and objective evaluation.

A Tone-Mapping Technique Based on Histogram Using a Sensitivity Model of the Human Visual System

Abstract: High-dynamic-range (HDR) images require tone mapping to be displayed properly on lower dynamic range devices. In this paper, a tone-mapping algorithm that uses histogram of luminance to construct a lookup table (LUT) for tone mapping is presented. Characteristics of the human visual system (HVS) are used to give more importance to visually distinguishable intensities while constructing the histogram bins. The method begins with constructing a histogram of the luminance channel, using bins that are perceived to be uniformly spaced by the HVS. Next, a refinement step is used, which removes the pixels from the bins that are indistinguishable by the HVS. Finally, the available display levels are distributed among the bins proportionate to the pixels counts thus giving due consideration to the visual contribution of each bin in the image. Quality assessment using both quantitative evaluations and user studies suggests that the presented algorithm produces tone-mapped images that are visually pleasant and preserve details of the original image better than the existing methods. Finally, implementation details of the algorithm on GPU for parallel processing are presented, which could achieve a significant gain in speed over CPU-based implementation.

Encoding perceptually-quantized video content in multi-layer vdr coding

Abstract: Input VDR images are received. A candidate set of function parameter values for a mapping function is selected from multiple candidate sets. A set of image blocks of non-zero standard deviations in VDR code words in at least one input VDR image is constructed. Mapped code values are generated by applying the mapping function with the candidate set of function parameter values to VDR code words in the set of image blocks in the at least one input VDR image. Based on the mapped code values, a subset of image blocks of standard deviations below a threshold value in mapped code words is determined as a subset of the set of image blocks. Based at least in part on the subset of image blocks, it is determined whether the candidate set of function parameter values is optimal for the mapping function to map the at least one input VDR image.

Retinex image enhancement via a learned dictionary

Abstract: The main aim of this paper is to study image enhancement by using sparse and redundant representations of the reflectance component in the Retinex model over a learned dictionary. This approach is different from existing variational methods, and the advantage of this approach is that the reflectance component in the Retinex model can be represented with more details by the dictionary. A variational method based on the dynamic dictionaries is adopted here, where it changes with respect to iterations of the enhancement algorithm. Numerical examples are also reported to demonstrate that the proposed methods can provide better visual quality of the enhanced high-contrast images than the other variational methods, i.e., revealing more details in the low-light part.

TMO-Net: A Parameter-Free Tone Mapping Operator Using Generative Adversarial Network, and Performance Benchmarking on Large Scale HDR Dataset

Abstract: Currently published tone mapping operators (TMO) are often evaluated on a very limited test set of high dynamic range (HDR) images. Thus, the resulting performance index is highly subject to extensive hyperparameter tuning, and many TMOs exhibit sub-optimal performance when tested on a broader spectrum of HDR images. This indicates that there are deficiencies in the generalizable applicability of these techniques. Finally, it is a challenge developing parameter-free tone mapping operators using data-hungry advanced deep learning methods due to the paucity of large scale HDR datasets. In this paper, these issues are addressed through the following contributions: a) a large scale HDR image benchmark dataset (LVZ-HDR dataset) with multiple variations in sceneries and lighting conditions is created to enable performance evaluation of TMOs across a diverse conditions and scenes that will also contribute to facilitate the development of more robust TMOs using state-of-the-art deep learning methods; b) a deep learning-based tone mapping operator (TMO-Net) is presented, which offers an efficient and parameter-free method capable of generalizing effectively across a wider spectrum of HDR content; c) finally, a comparative analysis, and performance benchmarking of 19 state-of-the-art TMOs on the new LVZ-HDR dataset are presented. Standard metrics including the Tone Mapping Quality Index (TMQI), Feature Similarity Index for Tone Mapped images (FSITM), and Natural Image Quality Evaluator (NIQE) are used to qualitatively evaluate the performance index of the benchmarked TMOs. Experimental results demonstrate that the proposed TMO-Net qualitatively and quantitatively outperforms current state-of-the-art TMOs.