• Journal of Korea Multimedia Society
• /
• v.19 no.8
• /
• pp.1310-1319
• /
• 2016

Face detection is the first step in a wide range of face applications. However, detecting faces in the wild is still a challenging task due to the wide range of variations in pose, scale, and occlusions. Recently, many deep learning methods have been proposed for face detection. However, further improvements are required in the wild. Another important issue to be considered in the face detection is the computational complexity. Current state-of-the-art deep learning methods require a large number of patches to deal with varying scales and the arbitrary image sizes, which result in an increased computational complexity. To reduce the complexity while achieving better detection accuracy, we propose a fully convolutional network-based face detection that can take arbitrarily-sized input and produce feature maps (heat maps) corresponding to the input image size. To deal with the various face scales, a multi-scale network architecture that utilizes the facial components when learning the feature maps is proposed. On top of it, we design multi-task learning technique to improve detection performance. Extensive experiments have been conducted on the FDDB dataset. The experimental results show that the proposed method outperforms state-of-the-art methods with the accuracy of 82.33% at 517 false alarms, while improving computational efficiency significantly.

• Proceedings of the IEEK Conference
• /
• 2000.11c
• /
• pp.145-148
• /
• 2000

This paper proposes a character segmentation method for Korean letter address image. The poor quality of image binarization results in broken character strokes. To overcome this problem, two steps of processing ate introduced. The first one is to merge broken characters to generate character candidates, and the other one is to reduce the complexity of segmentation graph path. These two steps do not use recognition information to keep in high-speed.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.22 no.8
• /
• pp.1715-1721
• /
• 1997

Wavelet transform is not applicable to arbitrarily-shaped region (or object) in images, due to the nature of its global decomposition. In this paper, the arbitrarily-shaped wavelet transform(ASWT) is proposed in order to solve this problem and its properties are investigated. Computation complexity of the ASWT is also examined and it is shown that the ASWT requires significantly fewer computations than conventional wavelet transform, since the ASWT processes only the object region in the original image. Experimental resutls show that any arbitrarily-shaped image segment can be decomposed using the ASWT and perfectly reconstructed using the inverse ASWT.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.5
• /
• pp.2273-2286
• /
• 2018

This paper describes transformations between elemental image arrays and a RGBD image for three-dimensional integral imaging and transmitting systems. Two transformations are introduced and analyzed in the proposed method. Normally, a RGBD image is utilized in efficient 3D data transmission although 3D imaging and display is restricted. Thus, a pixel-to-pixel mapping is required to obtain an elemental image array from a RGBD image. However, transformations and their analysis have little attention in computational integral imaging and transmission. Thus, in this paper, we introduce two different mapping methods that are called as the forward and backward mapping methods. Also, two mappings are analyzed and compared in terms of complexity and visual quality. In addition, a special condition, named as the hole-free condition in this paper, is proposed to understand the methods analytically. To verify our analysis, we carry out experiments for test images and the results indicate that the proposed methods and their analysis work in terms of the computational cost and visual quality.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.5
• /
• pp.601-608
• /
• 2014

In this study we propose an image denoising algorithm appropriate for mobile smart phone equipped with limited computing ability, which has better performance and at the same time comparable quality comparing with previous studies. The proposed image denoising algorithm for mobile smart phone cameras in low level light environment reduces computational complexity and also prevents edge smoothing by extracting just Gaussian noises from the noisy input image. According to the experiment result, we verified that our algorithm has much better PSNR value than methods applying mean filter or median filter. Also the result image from our algorithm has better clear quality since it preserves edges while smoothing input image. Moreover, the suggested algorithm reduces computational complexity about 52% compared to the method applying original Laplacian mask computation, and we verified that our algorithm has good denoising quality by implementing the algorithm in Android smart phone.

• Proceedings of the IEEK Conference
• /
• 2003.07e
• /
• pp.1899-1902
• /
• 2003

Recently several methods have been developed for the virtual space construction. Generally, most of the methods are geometric-based rendering technic, but they are difficult to construct real-time rendering because of large data. In this paper, we present a three dimension image-based rendering method that enable a constant speed of real-time rendering regardless of object complexity in virtual space. The Proposed method shows good performance for the virtual space construction with high complexity.

• Proceedings of the KIEE Conference
• /
• 1998.07g
• /
• pp.2225-2227
• /
• 1998

An efficient stereo matching method with variable matching weights is proposed. The edge segment-based stereo matching has been shown to be efficient method. The method includes 5 matching factor with different weights. The ordinary matching weights are not always adequate for every image. Employing different weight sets depending on the complexity shows better matching performance. In this paper, an evaluation criterion for complexity is suggested and the experimental results with the proposed method is shown.

• Proceedings of the IEEK Conference
• /
• 2001.09a
• /
• pp.121-124
• /
• 2001

For efficient transmission of 3DTV video signals, it is necessary to eliminate the inherent redundancy between the stereo image pairs. Though disparity estimation provides a powerful tool for eliminating the redundancy, it is very time consuming. This paper presents a novel disparity estimation scheme based on the human visual property. The disparity vectors of image blocks spatially adjacent to the current block are used as initial guesses fur the disparity vector of the current block. In addition, mixed-resolution coding is applied to reduce the computational complexity of disparity estimation. Through computer simulations on a stereoscopic sequence we show that the proposed method gives rise .to visually pleasing results with much reduced computational complexity.

• IEIE Transactions on Smart Processing and Computing
• /
• v.6 no.4
• /
• pp.292-298
• /
• 2017

The complexity of image compression algorithms has increased in order to improve image compression efficiency. One way to resolve high computational complexity is parallel processing. However, entropy coding, which is lossless compression, does not fit into the parallel processing form because of the correlation between consecutive symbols. This paper proposes a new application-specific instruction set processor (ASIP) platform by adding new context-adaptive binary arithmetic coding (CABAC) instructions to the existing platform to quickly process a variety of entropy coding. The newly added instructions work without conflicts with all other existing instructions of the platform, providing the flexibility to handle many coding standards with fast processing speeds. CABAC software is implemented for High Efficiency Video Coding (HEVC) and the performance of the proposed ASIP platform was verified with a field programmable gate array simulation.

• Proceedings of the KIEE Conference
• /
• 2007.04a
• /
• pp.304-306
• /
• 2007

Methods of Object recognition from camera image are to compare features of color. edge or pattern with model in a general way. SIFT(scale-invariant feature transform) has good performance but that has high complexity of computation. Using simple color histogram has low complexity. but low performance. In this paper we represent a model as a color cooccurrence histogram. and we improve performance using pyramid matching. The color cooccurrence histogram keeps track of the number of pairs of certain colored pixels that occur at certain separation distances in image space. The color cooccurrence histogram adds geometric information to the normal color histogram. We suggest object recognition by pyramid matching of color cooccurrence histogram.