• Korean Journal of Remote Sensing
• /
• v.35 no.6_1
• /
• pp.959-971
• /
• 2019

CNN (Convolutional Neural Network) is one representative deep learning algorithm, which can extract high-level spatial and spectral features, and has been applied for hyperspectral image classification. However, one significant drawback behind the application of CNNs in hyperspectral images is the high dimensionality of the data, which increases the training time and processing complexity. To address this problem, several CNN based hyperspectral image classification studies have exploited PCA (Principal Component Analysis) for dimensionality reduction. One limitation to this is that the spectral information of the original image can be lost through PCA. Although it is clear that the use of PCA affects the accuracy and the CNN training time, the impact of PCA for CNN based hyperspectral image classification has been understudied. The purpose of this study is to analyze the quantitative effect of PCA in CNN for hyperspectral image classification. The hyperspectral images were first transformed through PCA and applied into the CNN model by varying the size of the reduced dimensionality. In addition, 2D-CNN and 3D-CNN frameworks were applied to analyze the sensitivity of the PCA with respect to the convolution kernel in the model. Experimental results were evaluated based on classification accuracy, learning time, variance ratio, and training process. The size of the reduced dimensionality was the most efficient when the explained variance ratio recorded 99.7%~99.8%. Since the 3D kernel had higher classification accuracy in the original-CNN than the PCA-CNN in comparison to the 2D-CNN, the results revealed that the dimensionality reduction was relatively less effective in 3D kernel.

• Journal of Biomedical Engineering Research
• /
• v.40 no.5
• /
• pp.179-188
• /
• 2019

In ultrasound telemedicine, it is important to reduce the size of the data by compressing the ultrasound image when sending it. Ultrasound images can be divided into image context and other information consisting of patient ID, date, and several letters. Between them, ultrasound context is very important information for diagnosis and should be securely preserved as much as possible. In several previous papers, ultrasound compression methods were proposed to compress ultrasound context and other information into different compression parameters. This ultrasound compression method minimized the loss of ultrasound context while greatly compressing other information. This paper proposed the method of automatic segmentation of ultrasound context to overcome the limitation of the previously described ultrasound compression method. This algorithm was designed to robust for various ultrasound device and to enable real-time operation to maintain the benefits of ultrasound imaging machine. The operation time of extracting ultrasound context through the proposed segmentation method was measured, and it took 311.11 ms. In order to optimize the algorithm, the ultrasound context was segmented with down sampled input image. When the resolution of the input image was reduced by half, the computational time was 126.84 ms. When the resolution was reduced by one-third, it took 45.83 ms to segment the ultrasound context. As a result, we verified through experiments that the proposed method works in real time.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.12 no.4
• /
• pp.620-626
• /
• 2008

We propose a feature map method to detect a collision for a motion-based game. The feature map can be made an optimally reduced motion data using subtraction image and virtual ball images according to image size and condition. And we calculate the overlapped ratio between moving image data and objects. This ratio is an invariant for detection even though image size is changed. And we compare this ration with collision detection constant, the feature map can detect fast collisions as well as the collided direction. To evaluate the method, we implemented a motion-base game that consists of a web cam, a player, an enemy, and some virtual balls, and we obtained some valid results for our method for the collision detection. The results demonstrated that the proposed approach is robust, and they can be used as a basic collide detection algorithm for a motion-based game where the size and the position of characters are continuously changing.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.28B no.5
• /
• pp.318-326
• /
• 1991

A Quantizer reconstruction level control method for block artifact reduction in DCT image coding is described. In our scheme, quantizer reconstruction level control is obtained by adding quantization level step size to the optimum quantization level in the direction of reducing the block artifact by minimizing the mean square error(MSE) and error difference(EDF) distribution in boundary without the other additional bits. In simulation results, although the performance in terms of signal to noise ratio is degraded by a little amount, mean square of error difference at block boundary and mean square error having relation block artifact is greatly reduced. Subjective image qualities are improved compared with other block artifact reduction method such as postprocessing by filtering and trasform coding by block overlapping. But the addition calculations of 1-dimensional DCT become to be more necessary to coding process for determining the reconstruction level.

• Journal of Korea Multimedia Society
• /
• v.10 no.10
• /
• pp.1271-1283
• /
• 2007

To detect defect of FPD(flat panel displays) is very difficult due to uneven illumination on FPD panel image. This paper presents a method to detect various types of defects using the approximated image of the uneven illumination by B-spline. To construct a approximated surface, corresponding to uneven illumination background intensity, while reducing random noises and small defect signal, only the lowest smooth subband is used by wavelet decomposition, resulting in reducing the computation time of taking B-spline approximation and enhancing detection accuracy. The approximated image in lowest LL subband is expanded as the same size as original one by wavelet reconstruction, and the difference between original image and reconstructed one becomes a flat image of compensating the uneven illumination background. A simple binary thresholding is then used to separate the defective regions from the subtracted image. Finally, blob analysis as post-processing is carried out to get rid of false defects. For applying in-line system, the wavelet transform by lifting based fast algorithm is implemented to deal with a huge size data such as film and the processing time is highly reduced.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.40 no.1
• /
• pp.79-86
• /
• 2003

A new quantization constraint set based on the theory of Projection onto Convex Set(POCS) is proposed to reduce blocking artifact appearing in block-coded images. POCS-based postprocessing for alleviating the blocking artifact consists of iterative projections onto smoothness constraint set and quantization constraint set, respectively. In general, the conventional quantization constraint set has the maximum size of range where original image data can be included, therefore over-blurring of restored image is unavoidable as iteration proceeds. The projection onto the proposed quantization constraint set can reduce blocking artifact as well as maintain the clearness of the decoded image, since it controls adaptively the size of quantization constraint set according to the DCT coefficients. Simulation results using the proposed quantization constraint set as a substitute for conventional quantization constraint set show that the blocking artifact of the decoded image can be reduced by the small number of iterations, and we know that the postprocessed image maintains the distinction of the decoded image.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.37C no.11
• /
• pp.1045-1053
• /
• 2012

Link-16 is a widely used TDL (Tactical Data Link) which uses TDMA (Time Division Multiple Access). Link-16 is a very low rate system, so it supports small size of data like tactical message and voice. However, there are related works to transmit situation awareness information like image due to the increasing interest about EBO(Effect-Based Operation), recently. Special TDMA scheduling is needed not static TDMA of Link-16 for image transmission because image data has much larger size than the existing tactical data. In this paper, we proposed Link-16K which enhances the Link-16 MAC. The proposed Link-16K is compatible with Link-16, and includes dynamic TDMA, new packing method, and an efficient retransmission scheme for image transmission effectively. We can see that image transmission delay is reduced and channel utilization is increased through simulation results of proposed idea.

• Journal of Korea Multimedia Society
• /
• v.18 no.6
• /
• pp.701-709
• /
• 2015

Image stitching method is a technique for obtaining an high-resolution image by combining two or more images. In X-ray image for clinical diagnosis, the size of the imaging region taken by one shot is limited due to the field-of-view of the equipment. Therefore, in order to obtain a high-resolution image including large regions such as a whole body, the synthesis of multiple X-ray images is required. In this paper, we propose a rapid stitching method of digital X-ray images using template-based registration. The proposed algorithm use principal component analysis(PCA) and k-nearest neighborhood(k-NN) to determine the location of input images before performing a template-based matching. After detecting the overlapping position using template-based matching, we synthesize input images by alpha blending. To improve the computational efficiency, reduced images are used for PCA and k-NN analysis. Experimental results showed that our method was more accurate comparing with the previous method with the improvement of the registration speed. Our stitching method could be usefully applied into the stitching of 2D or 3D multiple images.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2001.05a
• /
• pp.265-268
• /
• 2001

In this paper, we designed one-dimensional VLSI array with high speed processing in Fractal image compression. fractal image compression algorithm partitions the original image into domain blocks and range blocks then compresses data using the self similarity of blocks. The image is partitioned into domain block with 50% overlapping. Domain block is reduced by averaging the original image to size of range block. VLSI array is trying to search the best matching between a range block and a large amount of domain blocks. Adjacent domain blocks are overlapped, so we can improve of each block's processing speed using the reuse of the overlapped data. In our experiment, proposed VLSI array has about 25% speed up by adding the least register, MUX, and DEMUX to the PE.

• The KIPS Transactions:PartB
• /
• v.10B no.1
• /
• pp.81-86
• /
• 2003

In this paper, we propose the tracking method of moving object which use extracted object by difference between background image and target image in fixed domain. As a extraction method of object, calculate not pixel of full image but predefined some edge pixel of image to get a position of new object. Since the center area Is excluded from calculation, the extraction time is efficiently reduced. To extract object in the predefined area, get a starting point in advance and then extract size of width and height of object. Central coordinate is used to track moved object.