• Journal of Korea Multimedia Society
• /
• v.22 no.11
• /
• pp.1223-1232
• /
• 2019

The injection is the input method of the image feature vector from the encoder to the decoder. Since the image feature vector contains object details such as color and texture, it is essential to generate image captions. However, the bidirectional decoder model using the existing injection method only inputs the image feature vector in the first step, so image feature vectors of the backward sequence are vanishing. This problem makes it difficult to describe the context in detail. Therefore, in this paper, we propose the parallel injection method to improve the description performance of image captions. The proposed Injection method fuses all embeddings and image vectors to preserve the context. Also, We optimize our image caption model with Bidirectional Gated Recurrent Unit (Bi-GRU) to reduce the amount of computation of the decoder. To validate the proposed model, experiments were conducted with a certified image caption dataset, demonstrating excellence in comparison with the latest models using BLEU and METEOR scores. The proposed model improved the BLEU score up to 20.2 points and the METEOR score up to 3.65 points compared to the existing caption model.

• 한국HCI학회:학술대회논문집
• /
• 2008.02a
• /
• pp.376-379
• /
• 2008

This paper presents an image retrieval algorithm based on feature vector using color of histogram refinement for a faster and more efficient search in the process of content based image retrieval. First, we segment each of R, G, and B images from RGB color image and extract their respective histograms. Secondly, these histograms of individual R, G and B are divided into sixteen of bins each. Finally, we extract the maximum pixel values in each bins' histogram, which are calculated, compared and analyzed, Now, we can perform image retrieval technique using these maximum pixel value. Hence, the proposed algorithm of this paper effectively extracts features by comparing input and database images, making features from R, G and B into a feature vector table, and prove a batter searching performance than the current algorithm that uses histogram matching and ranks, only.

• Journal of Korea Multimedia Society
• /
• v.2 no.3
• /
• pp.289-296
• /
• 1999

In this paper, an improved fractal color image decoding method using the data dependence parts and the vector distortion measure is proposed. The vector distortion measure exploits the correlation between different color components. The pixel in RGB color space can be considered as a 30dimensional vector with elements of RGB components. The root mean square error(rms) in RGB color for similarity measure of two blocks R and R' was used. We assume that various parameter necessary in image decoding are stored in the transform table. If the parameter is referenced in decoding image, then decoding is performed by the recursive decoding method. If the parameter is not referenced in decoding image, then the parameters recognize as the data dependence parts and store its in the memory. Non-referenced parts can be decoded only one time, because its domain informations exist in the decoded parts by the recursive decoding method. Non-referenced parts are defined the data dependence parts. Image decoding method using data dependence classifies referenced parts and non-referenced parts using information of transform table. And the proposed method can be decoded only one time for R region decoding speed than Zhang & Po's method, since it is decreased the computational numbers by execution iterated contractive transformations for the referenced range only.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.5
• /
• pp.42-47
• /
• 2017

In the distribution of digital image, the median filtering is used for a forgery. This paper proposed the algorithm of a image forensics detection for the classification of median filtering. For the solution of this grave problem, the feature vector is composed of 42-Dim. The detected quantity 32, 64 and 128 of forgery image edges, respectively, which are processed by the Hough transform, then it extracted from the start-end point coordinates of the Hough Lines. Also, the Hough Peaks of the Angle-Distance plane are extracted. Subsequently, both of the feature vectors are composed of the proposed scheme. The defined 42-Dim. feature vector is trained in SVM (Support Vector Machine) classifier for the MF classification of the forged images. The experimental results of the proposed MF detection algorithm is compared between the 10-Dim. MFR and the 686-Dim. SPAM. It confirmed that the MF forensic classification ratio of the evaluated performance is 99% above with the whole test image types: the unaltered, the average filtering ($3{\times}3$), the JPEG (QF=90 and 70)) compression, the Gaussian filtered ($3{\times}3$ and $5{\times}5$) images, respectively.

• Journal of KIISE:Databases
• /
• v.29 no.2
• /
• pp.128-137
• /
• 2002

In this paper, we propose the LPC+-file for efficient indexing of high-dimensional image data. With the proliferation of multimedia data, there Is an increasing need to support the indexing and retrieval of high-dimensional image data. Recently, the LPC-file (5) that based on vector approximation has been developed for indexing high-dimensional data. The LPC-file gives good performance especially when the dataset is uniformly distributed. However, compared with for the uniformly distributed dataset, its performance degrades when the dataset is clustered. We improve the performance of the LPC-file for the strongly clustered image dataset. The basic idea is to adaptively partition the data space to find subspaces with high-density clusters and to assign more bits to them than others to increase the discriminatory power of the approximation of vectors. The total number of bits used to represent vector approximations is rather less than that of the LPC-file since the partitioned cells in the LPC+-file share the bits. An empirical evaluation shows that the LPC+-file results in significant performance improvements for real image data sets which are strongly clustered.

• Proceedings of the Korean Information Science Society Conference
• /
• 2007.06c
• /
• pp.477-481
• /
• 2007

A background is a part where do not vary too much or frequently change in an image sequence. Using this assumption, it is presented a background acquisition algorithm for not only static but also dynamic view in this paper. For generating background, we detect a region, where has high correlation rate compared within selected region in the prior pyramid image, from the searching region in the current image. Between a detected region in the current image and a selected region in the prior image, we calculate movement vector for each regions in time sequence. After we calculate whole movement vectors for two successive images, vector histogram is used to determine the camera movement. The vector which has the highest density in the histogram is determined a camera movement. Using determined camera movement, we classify clusters based on pixel intensities which pixels are matched with prior pixels following camera movement. Finally we eliminate clusters which have lower weight than threshold, and combine remained clusters for each pixel to generate multiple background clusters. Experimental results show that we can automatically detect background whether camera move or not.

• Journal of KIISE:Information Networking
• /
• v.28 no.4
• /
• pp.651-656
• /
• 2001

Histogram intersection method, that counts the occurrence of color pixels, is one of the easy and simple color image retrieval methods. The method has an appropriate global property but does not contain the knowledge of shape for images. The absence of spatial information makes it difficult to discriminate images of the similar histogram. The application of one-dimensional projection to each image enables to obtain shape or spatial information of image. But in this case there is another problem having different length of the projection vector according to the size of each image. Thus this paper proposes a method that uses relative distances between peaks and their maximum value in the projection vector. In order to verify retrieval performance, the experimental results between the histogram intersection method, the projection only method, and the proposed one are compared and analyzed.

• KIEE International Transaction on Electrical Machinery and Energy Conversion Systems
• /
• v.5B no.4
• /
• pp.350-357
• /
• 2005

This paper presents a mechanical sensorless vector-controlled system with parameter identification by the aid of image processor. Based on the flux observer and the model reference adaptive system method, the proposed sensorless system includes rotor speed estimation and stator resistance identification using flux errors. Since the mathematical model of this system is constructed in a synchronously rotating reference frame, a linear model is easily derived for analyzing the system stability, including motor operating state and parameter variations. Because it is difficult to identify rotor resistance simultaneously while estimating rotor speed, a low-accuracy image processor is used to measure the mechanical axis position for calculating the rotor speed at a steady-state operation. The rotor resistance is identified by the error between the estimated speed using the estimated flux and the calculated speed using the image processor. Finally, the validity of this proposed system has been proven through experimentation.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.285-288
• /
• 2002

Compression of magnetic resonance images (MRI) has proved to be more difficult than other medical imaging modalities. In an average sized hospital, many tora bytes of digital imaging data (MRI) are generated every year, almost all of which has to be kept. The medical image compression is currently being performed by using different algorithms. In this paper, Fuzzy C-Means (FCM) algorithm is used for the Vector Quantization (VQ). First, a digital image is divided into subblocks of fixed size, which consists of 4${\times}$4 blocks of pixels. By performing 2-D Discrete Cosine Transform (DCT), we select six DCT coefficients to form the feature vector. And using FCM algorithm in constructing the VQ codebook. By doing so, the algorithm can make good time quality, and reduce the processing time while constructing the VQ codebook.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 1996.06a
• /
• pp.159-163
• /
• 1996

In this paper, a new image compression technique, which uses lifting scheme and pyramid vector quantization, is proposed. Lifting scheme is a new technique to generate wavelets and to perform wavelet transform, and pyramid vector quantization is a kind of vector quantization which dose not have codebook neither codebook generation algorithm. For the purpose of realizing more compression rate, an arithmetic entropy coder is used. Proposed algorithm is compared with other wavelet based image coder and with JPEG which uses DCT and adaptive Huffman entropy coder. Simulation results showed that the performance of proposed algorithm is much better than that of others in point of PSNR and bpp.