• Journal of Korea Multimedia Society
• /
• v.7 no.3
• /
• pp.319-326
• /
• 2004

In this paper, we propose color image segmentation by region merging method preserving the boundary of an object. The proposed method selects initial region by using quantized image's index map after vector quantizing an original image. After then, we merge regions by applying boundary restricted factor in order to consider the boundary of an object in HSI color space. Also we merge the regions in RGB color space for non-processed regions in HSI color space. And we reduce processing time by decreasing iterative process in region merging algorithm. Experimental results have demonstrated the superiority in region's segmentation results and processing time for various images.

• Proceedings of the KSRS Conference
• /
• v.2
• /
• pp.684-687
• /
• 2006

Airborne imagery must be precisely orthorectified to be used as geographical information data. GPS/INS (Global Positioning System/Inertial Navigation System) and LIDAR (LIght Detection And Ranging) data were employed to automatically orthorectify airborne images. In this study, 154 frame airborne images and LIDAR vector data were acquired. LIDAR vector data were converted to raster image for employing as reference data. To derive images with constant brightness, flat field correction was applied to the whole images. The airborne images were geometrically corrected by calculating internal orientation and external orientation using GPS/INS data and then orthorectified using LIDAR digital elevation model image. The precision of orthorectified images was validated using 50 ground control points collected in arbitrary selected five images and LIDAR intensity image. In validation results, RMSE (Root Mean Square Error) was 0.365 smaller then two times of pixel spatial resolution at the surface. It is possible that the derived mosaicked airborne image by this automatic orthorectification method is employed as geographical information data.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2003.06a
• /
• pp.94-97
• /
• 2003

In this Paper, propose for MLP(multilayer perception) neural network that uses optimization recognition training scheme for the wavelet transform and the numeral image add to noise, and apply this system in Numeral Recognition. As important part of original image information preserves maximum using the wavelet transform, node number of neural network and the loaming convergence time did size of input vector so that decrease. Apply in training vector, examine about change of the recognition rate as optimization recognition training scheme raises noise of data gradually. We used original image and original image added 0, 10, 20, 30, 40, 50㏈ noise (or the increase of numeral recognition rate. In case of test image added 30∼50㏈, numeral recognition rate between the original image and image added noise for training Is a little But, in case of test image added 0∼20㏈ noise, the image added 0, 10, 20, 30, 40 , 50㏈ noise is used training. Then numeral recognition rate improved 9 percent.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2001.10a
• /
• pp.579-582
• /
• 2001

Vector Transformation is a new method unified vector quantization and coding. So far, codebook generation applied to coding was LBG algorithm. But using the advantage of SOFM(Self-Organizing Feature Map) based on neural network can improve a system's performance. In this paper, we generated VTC(Vector Transformation Coding) codebook applied with SOFM algorithm and compare the result for several coding rates with LBG algorithm. The problem of Vector quantization is complicated calculation and codebook generation. So, to solve this problem, we used neural network approach method.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.37 no.2
• /
• pp.15-20
• /
• 2000

In this paper, we presents a new MCDI(motion compensated difference image) coding method using CVQ(classifled vector quantization) whoes MCD(motion compensated difference) block is classified by proposed classifier using motion vector and compensated block The variance of MCD block is closely related with the magnitude of motion vector as well as the variance of compensated block, so using this property, we propose a new classifier. This scheme has no side information of the classifier what sub-codebook is selected, and simulation results show that the proposed method exhibits a good performance even when compared with a conventional method that requires classification bits.

• The KIPS Transactions:PartB
• /
• v.11B no.1
• /
• pp.21-26
• /
• 2004

Color image quantization is a process of selecting a set of colors to display an image with some representative colors without noticeable perceived difference. It is very important in many applications to display a true color image in a low cost color monitor or printer. The basic problem is how to display 256 colors or less colors, called color palette, In this paper, we propose improved binary tree vector quantization based on spatial sensitivity which is one of the human visual properties. We combine the weights based on the responsibility of human visual system according to changes of three Primary colors in blocks of images with the process of splitting nodes using eigenvector in binary tree vector quantization. The test results show that the proposed method generates the quantized images with fine color and performs better than the conventional method in terms of clustering the similar regions. Also the proposed method can get the better result in subjective quality test and WSNR.

• Journal of Advanced Navigation Technology
• /
• v.24 no.5
• /
• pp.430-437
• /
• 2020

The proposed model implements a model that improves the face prediction rate and recognition rate through learning with an artificial neural network using face detection, landmark and face recognition algorithms. After landmarking in the face images of a specific person, the proposed model use the previously learned Caffe model to extract face detection and embedding vector 128D. The learning is learned by building machine learning algorithms such as support vector machine (SVM) and deep neural network (DNN). Face recognition is tested with a face image different from the learned figure using the learned model. As a result of the experiment, the result of learning with DNN rather than SVM showed better prediction rate and recognition rate. However, when the hidden layer of DNN is increased, the prediction rate increases but the recognition rate decreases. This is judged as overfitting caused by a small number of objects to be recognized. As a result of learning by adding a clear face image to the proposed model, it is confirmed that the result of high prediction rate and recognition rate can be obtained. This research will be able to obtain better recognition and prediction rates through effective deep learning establishment by utilizing more face image data.

• Journal of KIISE:Software and Applications
• /
• v.35 no.3
• /
• pp.162-168
• /
• 2008

A look-up table based vector adaptive filter is proposed in color inverse halftoning. Inverse halftoning converts halftone image into a continuous-tone image. The templates and training images are required in the process of look-up table based methods, which can be obtained from distributed patterns in the sample halftone images and their original images. Although the look-up table based methods usually are faster and show better performances in PSNR than other methods do, they show wide range of qualities depending on how they treat nonexisting patterns in the look-up table. In this paper, a vector adaptive filter is proposed to compensate for these nonexisting patterns, which achieves better quality owing to the contributed informations about hue, saturation, and intensity of surrounding pixels. The experimental results showed that the proposed method resulted in higher PSNR than that of conventional Best Linear Estimation method. The bigger the size of the template in the look-up table becomes, the more outstanding quality in the proposed method can be obtained.

• Journal of Korea Multimedia Society
• /
• v.9 no.11
• /
• pp.1465-1473
• /
• 2006

In this paper, we present an approach for facial expression recognition using Active Shape Models(ASM) and a state-based model in image sequences. Given an image frame, we use ASM to obtain the shape parameter vector of the model while we locate facial feature points. Then, we can obtain the shape parameter vector set for all the frames of an image sequence. This vector set is converted into a state vector which is one of the three states by the state-based model. In the classification step, we use the k-NN with the proposed similarity measure that is motivated on the observation that the variation-regions of an expression sequence are different from those of other expression sequences. In the experiment with the public database KCFD, we demonstrate that the proposed measure slightly outperforms the binary measure in which the recognition performance of the k-NN with the proposed measure and the existing binary measure show 89.1% and 86.2% respectively when k is 1.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.2S
• /
• pp.638-648
• /
• 2000

This paper introduces a new technique for block is matching motion estimation. Since the temporal correlation of the image sequence, the motion vector of a block is highly related to the motion vector of the same coordinate block in the previous image frame. If we can obtain useful and enough information from the motion vector of the same coordinate block of the previous frame, the total number of search points used to find the motion vector of the current block may be reduced significantly. Using that idea, an efficient predicted direction search algorithm (PDSA) for block matching algorithm is proposed. Based on the direction of the blocks of the two successive previous frames, if the direction of the to successive blocks is same, the first search point of the proposed PDSA is moved two pixels to the direction of the block. The searching process after moving the first search point is processed according to the fixed search patterns. Otherwise, full search is performed with search area $\pm$2. Simulation results show that PSNR values are improved up to the 3.4dB as depend on the image sequences and improved about 1.5dB on an average. Search times are reduced about 20% than the other fast search algorithms. Simulation results also show that the performance of the PDSA scheme gives better subjective picture quality than the other fast search algorithms and is closer to that of the FS(Full Search) algorithm.