• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.283-294
• /
• 2022

The feature map used in the network for deep learning generally has larger data than the image and a higher compression rate than the image compression rate is required to transmit the feature map. This paper proposes a method for transmitting a pyramid feature map with high compression rate, which is used in a network with an FPN structure that has robustness to object size in deep learning-based image processing. In order to efficiently compress the pyramid feature map, this paper proposes a structure that predicts a pyramid feature map of a level that is not transmitted with pyramid feature map of some levels that transmitted through the proposed prediction network to efficiently compress the pyramid feature map and restores compression damage through the proposed reconstruction network. Suggested mAP, the performance of object detection for the COCO data set 2017 Train images of the proposed method, showed a performance improvement of 31.25% in BD-rate compared to the result of compressing the feature map through VTM12.0 in the rate-precision graph, and compared to the method of performing compression through PCA and DeepCABAC, the BD-rate improved by 57.79%.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.318-331
• /
• 2022

With the recent development of hardware computing devices and software based frameworks, machine tasks using deep learning networks are expected to be utilized in various industrial fields and personal IoT devices. However, in order to overcome the limitations of high cost device for utilizing the deep learning network and that the user may not receive the results requested when only the machine task results are transmitted from the server, Collaborative Intelligence (CI) proposed the transmission of feature maps as a solution. In this paper, an efficient compression method for feature maps with vast data sizes to support the CI paradigm was analyzed and presented through experiments. This method increases redundancy by applying feature map reordering to improve compression efficiency in traditional video codecs, and proposes a feature map method that improves compression efficiency and maintains the performance of machine tasks by simultaneously utilizing image compression format and video compression format. As a result of the experiment, the proposed method shows 14.29% gain in BD-rate of BPP and mAP compared to the feature compression anchor of MPEG-VCM.

• Journal of Korea Multimedia Society
• /
• v.21 no.11
• /
• pp.1279-1287
• /
• 2018

In this paper, a new type of color coordinate conversion is proposed as modified CIEXYZ from RGB to compress the color gamut. The proposed segmentation includes principal component analysis for the optimal projection of a feature vector into a one-dimensional feature. The final step adopted for lip segmentation is Otsu's threshold for a two-class problem. The performance of the proposed method was better than that of conventional methods, especially for the chromatic feature.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.38 no.1
• /
• pp.11-18
• /
• 2001

In this paper, we proposed the image compression algorithm using triangular feature based GHA. In feature extraction, the input images are divided into eight areas of triangular shape, that has positional information for face image compression. The proposed algorithm reduces blocking effects in image reconstruction and contains informations of face feature and shapes of face as input images are divided into eight. We used triangular feature extraction for positional information and GHA for shape information of face images. Simulation results show that the proposed algorithm has a better performance than the block based K-means and non-parsed image based GHA in PSNR at the same bpp.

• Journal of Broadcast Engineering
• /
• v.27 no.1
• /
• pp.56-68
• /
• 2022

In this paper, we propose a compression method for multi-resolution feature maps for VCM. The proposed compression method removes the redundancy between the channels and resolution levels of the multi-resolution feature map through PCA-based transformation. According to each characteristic, the basis vectors and mean vector used for transformation, and the transformation coefficient obtained through the transformation are compressed using a VVC-based coder and DeepCABAC. In order to evaluate performance of the proposed method, the object detection performance was measured for the OpenImageV6 and COCO 2017 validation set, and the BD-rate of MPEG-VCM anchor and feature map compression anchor proposed in this paper was compared using bpp and mAP. As a result of the experiment, the proposed method shows a 25.71% BD-rate performance improvement compared to feature map compression anchor in OpenImageV6. Furthermore, for large objects of the COCO 2017 validation set, the BD-rate performance is improved by up to 43.72% compared to the MPEG-VCM anchor.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.53 no.6
• /
• pp.453-456
• /
• 2004

Due to the bandwidth and storage limitations medical images are needed to be compressed before transmission and storage. DICOM (Digital Imaging and Communications in Medicine) specification, which is the medical images standard, provides a mechanism for supporting the use of JPEG still image compression standard. In this paper, we explain a method for compressing medical images by PEG standard and propose two methods for JPEG compression. First, because medical images differ from natural images in optical feature, we propose a method to design adaptively the quantization table using spectrum analysis. Second, because medical images have higher pixel depth than natural images do, we propose a method to design Huffman table which considers the probability distribution feature of symbols. Simulation results show the improved performance compared to the quantization table and the adjusted Huffman table of JPEG standard.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.23 no.6
• /
• pp.519-526
• /
• 2013

ECG signal has the feature that is repeated in a cycle of P, Q, R, S, and T waves and is sampled at a high sampling frequency in general. By using the feature of periodic ECG signals, maximizing compression efficiency while minimizing the loss of important information for diagnosis is required. However, the periodic characteristics of such amplitude and period is not constant by measuring time and patients. Even though measured at the same time, the patient's characteristics display different periodic intervals. In this paper, an adaptive multi-level coding is provided by coding adaptively the dominant and non-dominant signal interval of the ECG signal. The proposed method can maximize the compression efficiency by using a multi-level code that applies different compression ratios considering information loss associated with the dominant signal intervals and non-dominant signal intervals. For the case of long time measurement, this method has a merit of maximizing compression ratio compared with existing compression methods that do not use the periodicity of the ECG signal and for the lossless compression coding of non-dominant signal intervals, the method has an advantage that can be stored without loss of information. The effectiveness of the ECG signal compression is proved throughout the experiment on ECG signal of MIT-BIH arrhythmia database.

• Journal of Multimedia Information System
• /
• v.6 no.2
• /
• pp.91-98
• /
• 2019

Many researchers have attempted to recognize three-dimensional faces using feature points extracted from two-dimensional facial photographs. However, due to the limit of flat photographs, it is very difficult to recognize faces rotated more than 15 degrees from original feature points extracted from the photographs. As such, it is difficult to create an algorithm to recognize faces in multiple angles. In this paper, it is proposed a new algorithm to recognize three-dimensional face recognition based on feature points extracted from a flat photograph. This method divides into six feature point vector zones on the face. Then, the vector value is compressed and expanded according to the rotation angle of the face to recognize the feature points of the face in a three-dimensional form. For this purpose, the average of the compressibility and the expansion rate of the face data of 100 persons by angle and face zone were obtained, and the face angle was estimated by calculating the distance between the middle of the forehead and the tail of the eye. As a result, very improved recognition performance was obtained at 30 degrees of rotated face angle.

• The Transactions of the Korea Information Processing Society
• /
• v.3 no.5
• /
• pp.1294-1306
• /
• 1996

To achieve high compression ratio for Hangul texts, in this paper two text compression algorithms which use the structural feature of Hangul, the frequency of postpositional words, are proposed. The performances of these proposed algorithms are also compared with previous text compression algorithms. The proposed compression algorithms named HLZ77 and HLZW come out from the modification of previous algorithms LZ77 and :ZW, respectively. The major distinction of the proposed ones is that the proposed algorithms use the fixed dictionary of selected postpositional words that appear most frequently in Hangul texts. The performances of HLZ77 and HLZW also are compared with those of LZ77 and LZW, respectively, with respect to the compression ratio. According to the result of performance study, the proposed algorithms are better than the previous algorithms for descriptive Hangul text snce the structural feature of Hangul is helpful to achievement of high compression ratio.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.295-307
• /
• 2022

In this paper, we propose a novel compression method for scalable Fisher vectors (SCFV) which is used as a global visual feature description of individual video frames in MPEG CDVA standard. CDVA standard has adopted a temporal descriptor redundancy removal technique that takes advantage of the correlation between global feature descriptors for adjacent keyframes. However, due to the variable length property of SCFV, the temporal redundancy removal scheme often results in inferior compression efficiency. It is even worse than the case when the SCFVs are not compressed at all. To enhance the compression efficiency, we propose an asymmetric SCFV difference computation method and a SCFV reconstruction method. Experiments on the FIVR dataset show that the proposed method significantly improves the compression efficiency compared to the original CDVA Experimental Model implementation.