Kwon Lee

Deinterlacing with motion adaptive vertical temporal filtering

In this paper, we propose a deinterlacing method with motion adaptive vertical temporal filtering, which utilizes the correlations between adjacent frames. In this model, we first interpolate the missing lines of the current frame and adjacent frames by using an intrafield deinterlacing method. Then we compute the pixel differences between the current frame and the adjacent frames. Since the differences between the adjacent frames would show similar patterns, we can use these patterns to improve deinterlacing performance. In other words, instead of performing deinterlacing in the frame domain, we perform the operation in the frame difference domain. Since the proposed method produces good performance in stationary regions, we selectively apply the vertical temporal filter. Then we apply the proposed method iteratively in order to enhance video quality. The proposed method shows low complexity and still produces superior performance. Experimental results show that the proposed method provides noticeable improvements over existing methods in terms of both subjective and objective evaluations.

Fast object detection based on color histograms and local binary patterns

In this paper, we present a fast object detection algorithm based on color histograms and local binary patterns (LBP). The proposed method consists of two steps: coarse target object detection and precise target object detection. During the coarse target object detection step, a small number of target candidates were generated using integral color histogram matching. To reduce computational complexity, we only used the dominant color of the integral color histogram matching. During the precise target object detection step, the target object was determined by comparing LBP feature distributions with those of the target object candidates. Experimental results show that the proposed method is efficient and robust.

No-reference video quality measurement using neural networks

Objective video quality measurements emerge as an important issue as multimedia data is increasingly transmitted over the channels where bandwidth may not be guaranteed. Among various objective models for video quality measurement, no-reference models have the largest application areas. In this paper, we propose a no-reference video quality assessment method for H.264 using artificial neural networks. Various features are extracted from H.264 bit-stream data and these features are inputted to a neural network. The neural network is trained to predict subjective video quality scores obtained by a number of evaluators. Experimental results show promising results, though a larger database would be required to train neural networks to provide robust performance.

High quality deinterlacing using content adaptive vertical temporal filtering

In this paper, we propose a content adaptive vertical temporal filtering method for deinterlacing which effectively uses correlations between adjacent frames. The proposed method consists of two steps: the initial interpolation step and the enhancement step. During the initial interpolation step, missing lines were reconstructed by applying a modified content adaptive vertical temporal filtering (MCAVTF) method. The MCAVTF method is proposed to improve the classification accuracy of the video content. After the initial interpolation step, the reconstructed lines were refined by using an adaptive weighted vertical temporal filtering (AWVTF) method. A field averaging filter was also used to enhance stationary local regions. Since we did not use motion estimation, the proposed method was of low complexity. Experimental results show that the proposed method outperforms existing methods in terms of subjective and objective evaluations.

M input radix p optical logic operations in a photorefractive BaTiO 3 crystal

A generalized coding algorithm for multiple inputs multiple-valued logic operations is proposed. This algorithm can reduce the unused pixels generated in the encoding process of the odd number of inputs. Using circle-type coded patterns and the correlation property in a photorefractive BaTiO(3) crystal, m input radix p logic operations are made possible. Detailed descriptions and advantages of the proposed method are presented and its effectiveness is demonstrated in the case of a three-input binary system. Uses for the proposed method, including binary half-adder/subtractor and binary multiplier, are given.

Content-based Image Retrieval using LBP and HSV Color Histogram

In this paper, we proposed a content-based image retrieval algorithm using local binary patterns and HSV color histogram. Images are retrieved using image input in image retrieval system. Many researches are based on global feature distribution such as color, texture and shape. These techniques decrease the retrieval performance in images which contained background the large amount of image. To overcome this drawback, the proposed method extract background fast and emphasize the feature of object by shrinking the background. The proposed method uses HSV color histogram and Local Binary Patterns. We also extract the Local Binary Patterns in quantized Hue domain. Experimental results show that the proposed method 82% precision using Corel 1000 database.

High quality spatially registered vertical temporal filtering for deinterlacing

In this paper, we propose a high quality deinterlacing method using vertical temporal filtering with spatial registration. Vertical temporal filtering methods generally perform well with low levels of complexity. However, they may also produce visible artifacts in moving scenes since incorrect blocks can be used from adjacent fields. By applying global spatial registration before performing vertical temporal filtering, we reduced these deinterlacing errors. To compute the global motion vector, we used a small number of pixels to reduce the computational complexity. However, this global spatial registration sometimes produced artifacts in stationary areas. To solve this problem, we selectively used the conventional vertical temporal filtering method and the spatially registered vertical temporal filtering method. We conducted experiments using CIF and HD progressive video sequences, some of which contained fast motion scenes. Experimental results show that the proposed method noticeably improved video quality in terms of subjective and objective evaluations. The proposed method showed 2-7 dB improvement in terms of PSNR compared to existing methods in fast moving video sequences.

Video quality measurement for multimedia applications using reduced-reference signals

As more multimedia services have become increasingly available over networks where bandwidth is not always guaranteed, quality monitoring has become an important issue. For instance, quality of experience and quality monitoring have become important problems in internet protocol television applications, since transmission errors may introduce all kinds of additional video quality degradations. In this paper, we present a reduced-reference objective model for video quality measurements in multimedia applications. The proposed method first measures edge degradations that are critical for perceptual video quality and then considers transmission error effects. We compared the proposed method with some existing methods. Independent verifications confirmed that the proposed method showed good performance and consequently it was included in an International Telecommunication Union recommendation. The proposed method can be used to monitor video quality at receivers while requiring minimum usage of additional bandwidth.

Effective Deinterlacing Using Selective Spatial-Temporal Interpolation

In this paper, we propose a new deinterlacing algorithm, which selectively uses spatial and temporal interpolations. Conventional deinterlacing methods tend to yield undesired results in horizontal edges and high frequency areas. To address these problems, the proposed method uses different interpolation methods based on motion types. We first apply motion estimation and intrafield interpolation in the spatial domain. Second, a new interfield interpolation technique is performed. Finally, we selectively apply motion-compensation deinterlacing or repeat the new interfield interpolation technique. Experimental results show that the proposed algorithm provides noticeably improved performance compared to existing deinterlacing algorithms.

Octagonal prism LBP representation for face recognition

In this paper, we propose an octagonal prism representation for local binary patterns (LBP). This representation implements a new circular distance measurement for face recognition under various illumination conditions. The LBP method has been widely used in many computer vision applications, particularly for face recognition. Most LBP matching methods use distribution features with a bin-to-bin distance measure. However, using this bin-to-bin distance measure may produce low similarity scores even for similar patterns. To address this problem, we placed the LBPs on an octagonal prism in a three dimensional space and used the Euclidean distance measure. In the proposed octagonal prism representation, the LBPs were represented as three dimensional vectors on the octagonal prism. Since similar patterns under different illumination conditions are located in the vicinity on the octagonal prism, the proposed method proved robust against illumination variations. The proposed method produced noticeably improved performance when using the CMU PIE, Yale B, and Extended Yale B databases.