• Journal of KIISE:Software and Applications
• /
• v.31 no.2
• /
• pp.189-202
• /
• 2004

In this paper, we propose a new salient region detection method in an image. The algorithm is based on the characteristics of human's bottom-up visual attention. Several features known to influence human visual attention like color, intensity and etc. are extracted from the each regions of an image. These features are then converted to importance values for each region using its local competition function and are combined to produce a saliency map, which represents the saliency at every location in the image by a scalar quantity, and guides the selection of attended locations, based on the spatial distribution of saliency region of the image in relation to its Perceptual importance. Results shown indicate that the calculated Saliency Maps correlate well with human perception of visually important regions.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.8
• /
• pp.123-132
• /
• 2017

Traffic lights contain very important information for safety driving. So, the delivery of the information to drivers in real-time is a very critical issue for advanced driver assistance systems. However, traffic light detection is quite difficult because of the small sized traffic lights and the occlusion in real world. In this paper, a traffic light detection method using modified color based saliency map and morphological information is proposed. It shows 98.14% of precisions and 83.52% of recalls on computer simulations.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.3
• /
• pp.1264-1286
• /
• 2018

In this paper, a novel background prior-based salient object detection framework is proposed to deal with images those are more complicated. We take the superpixels located in four borders into consideration and exploit a mechanism based on image boundary information to remove the foreground noises, which are used to form the background prior. Afterward, an initial foreground prior is obtained by selecting superpixels that are the most dissimilar to the background prior. To determine the regions of foreground and background based on the prior of them, a threshold is needed in this process. According to a fixed threshold, the remaining superpixels are iteratively assigned based on their proximity to the foreground or background prior. As the threshold changes, different foreground priors generate multiple different partitions that are assigned a likelihood of being foreground. Last, all segments are combined into a saliency map based on the idea of similarity voting. Experiments on five benchmark databases demonstrate the proposed method performs well when it compares with the state-of-the-art methods in terms of accuracy and robustness.

• Journal of Korea Multimedia Society
• /
• v.20 no.5
• /
• pp.769-781
• /
• 2017

Deep learning networks like Convolutional Neural Networks (CNNs) show successful performances in many computer vision applications such as image classification, object detection, and so on. For implementation of deep learning networks in embedded system with limited processing power and memory, deep learning network may need to be simplified. However, simplified deep learning network cannot learn every possible scene. One realistic strategy for embedded deep learning network is to construct a simplified deep learning network model optimized for the scene images of the installation place. Then, automatic training will be necessitated for commercialization. In this paper, as an intermediate step toward automatic training under fisheye camera environments, we study more precise human localization in fisheye images, and propose an accurate human localization method, Automatic Ground-Truth Labelling Method (AGTLM). AGTLM first localizes candidate human object bounding boxes by utilizing GoogLeNet-LSTM approach, and after reassurance process by GoogLeNet-based CNN network, finally refines them more correctly and precisely(tightly) by applying saliency object detection technique. The performance improvement of the proposed human localization method, AGTLM with respect to accuracy and tightness is shown through several experiments.

• Journal of the Korea Society of Computer and Information
• /
• v.19 no.1
• /
• pp.9-21
• /
• 2014

In this paper, we propose the auto correction technique of photography composition by which the eye line is concentrated and the stable image of the structure can be obtained in case the general user takes a picture. Because the general user photographs in most case without background knowledge about the composition of the photo, the subject location is not appropriate and the unstable composition is contrasted with the stable composition of pictures which the experts take. Therefore, we provide not the method processing the image after photographing, but he method presenting automatically the stable composition when the general users take a photograph. The proposed method analyze the subject through Saliency Map, Image Segmentation, Edge Detection, etc. and outputs the subject at the location where the stable composition can be comprised along with the guideline of the Rule of Thirds. The experimental result shows that the good composition was presented to the user automatically.

• Journal of Information Technology and Architecture
• /
• v.10 no.1
• /
• pp.113-122
• /
• 2013

In this paper, we propose the method that automatically corrects the composition of a picture stylishly as well as reliably by cropping pictures based on the Rule of Thirds. The region of interest (ROI) is extracted from a picture by applying the Saliency Map and the Image Segmentation technology, the composition of the photo is amended based on this area to satisfy the Rule of Thirds. In addition, since the face region of the person is added to ROI by the Face Detection technique and the composition is amended by the various scenario according to ROI, the little more natural picture is acquired. The experimental result shows that the photo of the corrected composition was naturally amended compared with the original photo.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.29 no.1
• /
• pp.117-125
• /
• 2019

Deep Learning, which is being used in various fields recently, is being threatened with Adversarial Attack. In this paper, we experimentally verify that the classification accuracy is lowered by adversarial samples generated by malicious attackers in image classification models. We used MNIST dataset and measured the detection accuracy by injecting adversarial samples into the Autoencoder classification model and the CNN (Convolution neural network) classification model, which are created using the Tensorflow library and the Pytorch library. Adversarial samples were generated by transforming MNIST test dataset with JSMA(Jacobian-based Saliency Map Attack) and FGSM(Fast Gradient Sign Method). When injected into the classification model, detection accuracy decreased by at least 21.82% up to 39.08%.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2016.06a
• /
• pp.33-34
• /
• 2016

본 논문에서는 HVS(human visual system)의 특성을 고려한 새로운 스케일러블 코딩방법을 제안한다. 제안된 방법은 먼저 영상 내에서 관심영역(saliency map)을 찾고 관심영역을 제외한 부분에 에지 보존 필터를 적용한다. 그 영상은 정해진 양자 파라미터 값으로 인코딩 되어 제안된 코딩 시스템의 베이스 층(base layer)이 된다. 기존 스케일러블 코딩 표준에서의 베이스 층과 다르게 본 논문의 베이스 층은 관심 있는 중요영역(foreground)을 보존하고 또한 배경(background)의 에지 성분도 보존한다. 기본 층이 전송되면 개선층(enhancement layer)은 원 영상과 복원된 베이스 층 영상간의 차분 영상에서 관심영역 순으로 보내진다. 실험은 HEVC 를 바탕으로 수행되었고 스케일러블 코딩 표준인 SHVC 와 관심영역에서 비교를 했을 때 제안된 알고리즘이 더 높은 PSNR 을 가지는 것을 확인하였다. 또한 전체적으로 지각적인 품질(perceptual quality) 또한 향상되었음을 확인하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2011.04a
• /
• pp.384-385
• /
• 2011

본 논문은 중요도 맵과 Mean Shift 알고리즘을 이용하여 모바일 폰 영상 내의 와인 라벨 검출 방법을 제안한다. Mean Shift 알고리즘은 비모수적 클러스터링 기술로 클러스터의 수에 대한 사전 지식이 없이도 클러스터링이 가능한 알고리즘인데 실행 시간이 많이 필요한 단점이 있다. 이러한 문제를 해결하기 위해서 입력 칼라 와인 영상에 Saliency Map을 먼저 적용하고 영상의 두드러진 영역을 찾는다. 다음으로 Mean Shift 알고리즘을 이용한 분할 결과에서 얻은 칼라 마스크를 따라 빈도가 가장 높은 칼라 영역을 찾고 와인 라벨 영역을 검출한다. 실험결과를 통하여 제안된 방법을 모바일 폰을 이용하여 획득된 다양한 와인 영상의 라벨 영역을 효율적으로 검출할 수 있음을 볼 수 있다.

• Journal of Broadcast Engineering
• /
• v.17 no.5
• /
• pp.838-850
• /
• 2012

In this paper, we propose a novel algorithm that generates multi-view images by using various image features obtained from the given stereoscopic images. In the proposed algorithm, we first create an intensity gradient saliency map from the given stereo images. And then we calculate a block-based optical flow that represents the relative movement(disparity) of each block with certain size between left and right images. And we also obtain the disparities of feature points that are extracted by SIFT(scale-invariant We then create a disparity saliency map by combining these extracted disparity features. Disparity saliency map is refined through the occlusion detection and removal of false disparities. Thirdly, we extract straight line segments in order to minimize the distortion of straight lines during the image warping. Finally, we generate multi-view images by grid mesh-based image warping algorithm. Extracted image features are used as constraints during grid mesh-based image warping. The experimental results show that the proposed algorithm performs better than the conventional DIBR algorithm in terms of visual quality.