• Journal of Korea Multimedia Society
• /
• v.19 no.12
• /
• pp.1909-1918
• /
• 2016

This paper presents a unified framework for joint Convolutional Neural Network (CNN) based vehicle detection by leveraging multi-spectral image pairs. With the observation that under challenging environments such as night vision and limited light source, vehicle detection in a single color image can be more tractable by using additional far-infrared (FIR) image, we design joint CNN architecture for both RGB and FIR image pairs. We assume that a score map from joint CNN applied to overall image can be considered as confidence of vehicle existence. To deal with various scale ratios of vehicle candidates, multi-scale images are first generated scaling an image according to possible scale ratio of vehicles. The vehicle candidates are then detected on local maximal on each score maps. The generation of overlapped candidates is prevented with non-maximal suppression on multi-scale score maps. The experimental results show that our framework have superior performance than conventional methods with a joint framework of multi-spectral image pairs reducing false positive generated by conventional vehicle detection framework using only single color image.

• Journal of Electrical Engineering and Technology
• /
• v.10 no.4
• /
• pp.1886-1898
• /
• 2015

In today’s era of advanced technological developments, the threats to the authenticity and integrity of digital images, in a nutshell, the threats to the Image Forensics Research communities have also increased proportionately. This happened as even for the ‘non-expert’ forgers, the availability of image processing tools has become a cakewalk. This image forgery poses a great problem for judicial authorities in any context of trade and commerce. Block matching based image cloning detection system is widely researched over the last 2-3 decades but this was discouraged by higher computational complexity and more time requirement at the algorithm level. Thus, for reducing time need, various dimension reduction techniques have been employed. Since a single technique cannot cope up with all the transformations like addition of noise, blurring, intensity variation, etc. we employ multiple techniques to a single image. In this paper, we have used Fuzzy logic approach for decision making and getting a global response of all the techniques, since their individual outputs depend on various parameters. Experimental results have given enthusiastic elicitations as regards various transformations to the digital image. Hence this paper proposes Fuzzy based cloning detection and classification system. Experimental results have shown that our detection system achieves classification accuracy of 94.12%. Detection accuracy (DAR) while in case of 81×81 sized copied portion the maximum accuracy achieved is 99.17% as regards subjection to transformations like Blurring, Intensity Variation and Gaussian Noise Addition.

• Journal of Multimedia Information System
• /
• v.9 no.2
• /
• pp.75-86
• /
• 2022

The existing image quality assessment (IQA) datasets have a small number of samples. Some methods based on transfer learning or data augmentation cannot make good use of image quality-related features. A No Reference (NR)-IQA method based on multi-task training and quality awareness is proposed. First, single or multiple distortion types and levels are imposed on the original image, and different strategies are used to augment different types of distortion datasets. With the idea of weak supervision, we use the Full Reference (FR)-IQA methods to obtain the pseudo-score label of the generated image. Then, we combine the classification information of the distortion type, level, and the information of the image quality score. The ResNet50 network is trained in the pre-train stage on the augmented dataset to obtain more quality-aware pre-training weights. Finally, the fine-tuning stage training is performed on the target IQA dataset using the quality-aware weights to predicate the final prediction score. Various experiments designed on the synthetic distortions and authentic distortions datasets (LIVE, CSIQ, TID2013, LIVEC, KonIQ-10K) prove that the proposed method can utilize the image quality-related features better than the method using only single-task training. The extracted quality-aware features improve the accuracy of the model.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.43 no.1 s.307
• /
• pp.45-52
• /
• 2006

In this paper, a real-time stereoscopic image conversion method using a single frame from a 2-D image is proposed. The Stereoscopic image is generated by creating depth map using vortical position information and parallax processing. For a real-time processing of stereoscopic conversion and reduction of hardware complexity, it uses image sampling, object segmentation by standardizing luminance and depth map generation by boundary scan. The proposed method offers realistic 3-D effect regardless of the direction, velocity and scene conversion of the 2-D image. It offers effective stereoscopic conversion using images suitable conditions assumed in this paper such as recorded image at long distance, landscape and panorama photo because it creates different depth sense using vertical position information from a single frame. The proposed method can be applied to still image because it uses a single frame from a 2-D image. The proposed method has been evaluated using visual test and APD for comparing the stereoscopic image of the proposed method with that of MTD. It is confirmed that stereoscopic images conversed by the proposed method offers 3-D effect regardless of the direction and velocity of the 2-D image.

• Journal of Korean Society for Geospatial Information Science
• /
• v.17 no.3
• /
• pp.71-79
• /
• 2009

We expanded the 3D building information extraction method using shadow and vertical line from single high resolution image with meta information into the method for single high resolution image without meta information. Our method guesses an azimuth angle and an elevation angle of the sensor and the sun using reference building, selected by user, on an image. For test, we used an IKONOS image and an image extracted from the Google Earth. We calculated the Root Mean Square (RMS) error of heights extracted by our method using the building height extracted from stereo IKONOS image as reference, and the RMS error from the IKONOS image and the Google Earth image was under than 3 m. We also calculated the RMS error of horizontality position by comparison between building position extracted from only the IKONOS image and it from 1:1,000 digital map, and the result was under than 3 m. This test results showed that the height pattern of building models by our method was similar with it by the method using meta information.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.5
• /
• pp.1814-1828
• /
• 2021

Low-light image enhancement is a key technique to overcome the quality degradation of photos taken under scotopic vision illumination conditions. The degradation includes low brightness, low contrast, and outstanding noise, which would seriously affect the vision of the human eye recognition ability and subsequent image processing. In this paper, we propose an approach based on deep learning and Retinex theory to enhance the low-light image, which includes image decomposition, illumination prediction, image reconstruction, and image optimization. The first three parts can reconstruct the enhanced image that suffers from low-resolution. To reduce the noise of the enhanced image and improve the image quality, a super-resolution algorithm based on the Laplacian pyramid network is introduced to optimize the image. The Laplacian pyramid network can improve the resolution of the enhanced image through multiple feature extraction and deconvolution operations. Furthermore, a combination loss function is explored in the network training stage to improve the efficiency of the algorithm. Extensive experiments and comprehensive evaluations demonstrate the strength of the proposed method, the result is closer to the real-world scene in lightness, color, and details. Besides, experiments also demonstrate that the proposed method with the single low-light image can achieve the same effect as multi-exposure image fusion algorithm and no ghost is introduced.

• Journal of Broadcast Engineering
• /
• v.13 no.1
• /
• pp.162-173
• /
• 2008

In this paper, we propose an effective stereo-view video coding method that considers stereo-view and displayer characteristics. Current many stereo video displayers are designed for not only stereo display but also conventional single view display. In these systems, the resolution of two input videos for a stereo mode is half of that of single view for compatibility with conventional single view video services. In this raper, we propose a stereo video codec to deal with both single view and stereo view services by encoding whole left image and down-sampled right image. However, direct disparity estimation is not possible between two views because the resolution of a left image is different from that of the corresponding right image. So, we propose a disparity estimation method to make use of full information of the left reference image without down-sampling. In experimental result, we achieved $0.5{\sim}0.8\;dB$ coding gain, compared with several conventional algorithms.

• Journal of Broadcast Engineering
• /
• v.24 no.2
• /
• pp.281-291
• /
• 2019

Depth from defocus estimates the 3D depth by using a phenomenon in which the object in the focal plane of the camera forms a clear image but the object away from the focal plane produces a blurred image. In this paper, algorithms are studied to estimate 3D depth by analyzing the degree of blur of the image taken with a single camera. The optimized object range was obtained by 3D depth estimation derived from depth from defocus using one image of a single camera or two images of different focus of a single camera. For depth estimation using one image, the best performance was achieved using a focal length of 250 mm for both smartphone and DSLR cameras. The depth estimation using two images showed the best 3D depth estimation range when the focal length was set to 150 mm and 250 mm for smartphone camera images and 200 mm and 300 mm for DSLR camera images.

• Proceedings of the Korean Society for Technology of Plasticity Conference
• /
• 2007.05a
• /
• pp.51-54
• /
• 2007

The purpose of this paper is the development of measurement algorithm for green-sheet based on the digital image processing technique. The Low Temperature Cofired Ceramic (LTCC) technology can be defined as a way to produce multilayer circuits with the help of single tapes, which are used to apply conductive, dielectric and / or resistive pastes on. These single green-sheets have to be laminated together and fired in one step all. Main functionality of the green-sheet film measurement algorithm is to measure the position and size of the punching hole in each single layer. The line scan camera coupled with motorized X-Y stage is used for developing the algorithm. In order to measure the entire film area using several scanning steps, the overlapping method is used. In the process of development of the algorithm based on the image processing and analysis, strong background technology and know-how have been accumulated.

• The Journal of Korea Robotics Society
• /
• v.15 no.4
• /
• pp.385-391
• /
• 2020

As deep learning technology has been developed and applied to various fields, it is gradually changing from an existing single image based application to a video based application having a time base in order to recognize human behavior. However, unlike 2D CNN in a single image, 3D CNN in a video has a very high amount of computation and parameter increase due to the addition of a time axis, so improving accuracy in action recognition technology is more difficult than in a single image. To solve this problem, we investigate and analyze various techniques to improve performance in 3D CNN-based image recognition without additional training time and parameter increase. We propose a time base ensemble using the time axis that exists only in the videos and an ensemble in the input frame. We have achieved an accuracy improvement of up to 7.1% compared to the existing performance with a combination of techniques. It also revealed the trade-off relationship between computational and accuracy.