• Journal of the Korea Computer Graphics Society
• /
• v.27 no.3
• /
• pp.1-11
• /
• 2021

Slow sync is a photography technique where a user takes an image with long exposure and a camera flash to enlighten the foreground and background. Unlike short exposure with flash and long exposure without flash, slow sync guarantees the bright foreground and background in the dim environment. However, taking a slow sync image with a smartphone is difficult because the smartphone camera has continuous and weak flash and can not turn on flash if the exposure time is long. This paper proposes a deep learning method that input is a short exposure flash image and output is a slow sync image. We present a deep learning network with a weight map for spatially varying enlightenment. We also propose a dataset that consists of smartphone short exposure flash images and slow sync images for supervised learning. We utilize the linearity of a RAW image to synthesize a slow sync image from short exposure flash and long exposure no-flash images. Experimental results show that our method trained with our dataset synthesizes slow sync images effectively.

• 한국정보디스플레이학회:학술대회논문집
• /
• 2005.07b
• /
• pp.1615-1618
• /
• 2005

In this paper, implementation of object-based multiview 3D display using object segmentation and adaptive disparity estimation is proposed and its performance is analyzed by comparison to that of the conventional disparity estimation algorithms. In the proposed algorithm, firstly we can get segmented objects by region growing from input stereoscopic image pair and then, in order to effectively synthesize the intermediate view the matching window size is selected according to the extracted feature value of the input stereo image pair. Also, the matching window size for the intermediate view reconstruction (IVR) is adaptively selected in accordance with the magnitude of the extracted feature value from the input stereo image pair. In addition, some experimental results on the IVR using the proposed algorithm is also discussed and compared with that of the conventional algorithms.

• Journal of Digital Contents Society
• /
• v.19 no.1
• /
• pp.205-212
• /
• 2018

In this paper, we propose a method for more accurate fake discrimination using time information in CNN-based signature recognition. To easily use the time information and not to be influenced by the speed of signature writing, we acquire the signature as a movie and divide the total time of the signature into equal numbers of equally spaced intervals to obtain each image and synthesize them to create signature data. In order to compare the method using the proposed signature image and the method using only the last signature image, various signature recognition methods based on CNN have been experimented in this paper. As a result of experiment with 25 signature data, we found that the method using time information improves performance in fake discrimination compared to the existing method at all experiments.

• Journal of Internet Computing and Services
• /
• v.22 no.3
• /
• pp.37-43
• /
• 2021

In this paper, we study how objects from front and back cameras can be recognized in real time in a mobile environment to segment regions of object pixels and synthesize them through image processing. To this work, we applied DeepLabV3 machine learning model to dual cameras provided by Apple's iOS. We also propose methods using Core Image and Core Graphics libraries from Apple for image synthesis and postprocessing. Furthermore, we improved CPU usage than previous works and compared the throughput rates and results of Depth and DeepLabV3. Finally, We also developed a camera application using these two methods.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.8
• /
• pp.1136-1141
• /
• 2022

Researches of face recognition on masked faces have been increasingly important due to the COVID-19 pandemic. To realize a stable and practical recognition performance, large amount of facial image data should be acquired for the purpose of training. However, it is difficult for the researchers to obtain masked face images for each human subject. This paper proposes a novel method to synthesize a face image and a virtual mask pattern. In this method, a pair of masked face image and unmasked face image, that are from a single human subject, is fed into a convolutional autoencoder as training data. This allows learning the geometric relationship between face and mask. In the inference step, for a unseen face image, the learned convolutional autoencoder generates a synthetic face image with a mask pattern. The proposed method is able to rapidly generate realistic masked face images. Also, it could be practical when compared to methods which rely on facial feature point detection.

• Journal of the HCI Society of Korea
• /
• v.2 no.1
• /
• pp.25-31
• /
• 2007

We develop an interactive relighting renderer allowing camera view changes based on a deep-frame buffer approach. The renderer first caches the rendering parameters for a given 3D scene in an auxiliary buffer with the same size of the output image. The rendering parameters independent from light changes are selected from the shading models used for shading pixels. Next, as the user interactively edits one light at one time, the relighting renderer instantly re-shades each pixel by updating the contribution of the changed light with the shading parameters cached in the deep-frame buffer. When the camera moves, the cache values should be re-computed because the currently cached values become obsolete. We present a novel method to synthesize them quickly from the cache images of the user specified cameras by using an image-based technique. This computations are all performed on GPU to achieve real-time performance.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.20 no.3
• /
• pp.630-638
• /
• 2016

In this paper, a feature based panorama image generation algorithm using FAST(Features from Accelerated Segment Test) method that is faster than SIFT(Scale Invariant Feature Transform) and SURF(Speeded Up Robust Features) is proposed. Cylindrical projection is performed to generate natural panorama images with numerous images as input. The occurred error can be minimized by applying RANSAC(Random Sample Consensus) for the matching process. When we synthesize numerous images acquired from different camera angles, we use blending techniques to compensate the distortions by the heterogeneity of border line. In that way, we could get more natural synthesized panorama image. The proposed algorithm can generate natural panorama images regardless the order of input images and tilted images. In addition, the image matching can be faster than the conventional method. As a result of the experiments, distortion was corrected and natural panorama image was generated.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.12
• /
• pp.2946-2952
• /
• 2014

Among the Example based Super Resolution(SR) techniques, Neighbor embedding(NE) has been inspired by manifold learning method, particularly locally linear embedding. However, the poor generalization of NE decreases the performance of such algorithm. The sizes of local training sets are always too small to improve the performance of NE. We propose the Learning Sparse-Neighbor Image Representation baesd on SVR having an excellent generalization ability to solve this problem. Given a low resolution image, we first use bicubic interpolation to synthesize its high resolution version. We extract the patches from this synthesized image and determine whether each patch corresponds to regions with high or low spatial frequencies. After the weight of each patch is obtained by our method, we used to learn separate SVR models. Finally, we update the pixel values using the previously learned SVRs. Through experimental results, we quantitatively and qualitatively confirm the improved results of the proposed algorithm when comparing with conventional interpolation methods and NE.

• Journal of the Korea Computer Graphics Society
• /
• v.13 no.4
• /
• pp.7-12
• /
• 2007

With the advent of digital cinema, more and more movies are digitally produced, distributed via digital medium such as hard drives and network, and finally projected using a digital projector. However, digital cameras capable of shotting at 2K or higher resolution for digital cinema are still very expensive and bulky, which impedes rapid transition to digital production. As a low-cost solution for acquiring high resolution digital videos, we propose a hybrid camera consisting of a low-resolution CCD for capturing videos and a high-resolution CCD for capturing still images at regular intervals. From the output of the hybrid camera, we can synthesize high-resolution videos by software as follows: for each frame, 1. find pixel correspondences from the current frame to the previous and subsequent keyframes associated with high resolution still images, 2. synthesize a high-resolution image for the current frame by copying the image blocks associated with the corresponding pixels from the high-resolution keyframe images, and 3. complete the synthesis by filling holes in the synthesized image. This framework can be extended to making NPR video effects and capturing HDR videos.

• Journal of KIISE:Software and Applications
• /
• v.27 no.6
• /
• pp.618-631
• /
• 2000

In this paper, we suggest the method of 3D facial synthesis using the motion of 2D facial images. We use the optical flow-based method for estimation of motion. We extract parameterized motion vectors using optical flow between two adjacent image sequences in order to estimate the facial features and the facial motion in 2D image sequences. Then, we combine parameters of the parameterized motion vectors and estimate facial motion information. We use the parameterized vector model according to the facial features. Our motion vector models are eye area, lip-eyebrow area, and face area. Combining 2D facial motion information with 3D facial model action unit, we synthesize the 3D facial model.