Hideaki Kimata

Hideaki Kimata

Auto-generated profile, based on publicly available information

Hideaki Kimata

Department of Informatics, Kogakuin University, 1-24-2 Nishi-Shinjuku, Shinjuku-ku, 163-8677, Tokyo, Japan

Research Focus & Metrics

Academic Career Length

16Years (2011-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 August 2021 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
SSSatoshi Suzuki
Satoshi Suzuki
STShoichiro Takeda
Shoichiro Takeda
MTMotohiro Takagi
Motohiro Takagi
RTRyuichi Tanida
Ryuichi Tanida
HKHideaki Kimata
Hideaki Kimata
HSHayaru Shouno
Hayaru Shouno
Collaborative Intelligence is a new paradigm that splits a deep neural network (DNN) into an edge and cloud for deploying a DNN-based image recognition application. In this paradigm, deep features, which are the outputs of the edge DNN, are compressed and transmitted to the cloud DNN. Because the deep features have a number of responses that are similar to each other, for efficient compression, previous methods spatially arrange and compress the deep features as an image to utilize the similarity as a spatial correlation. However, if the deep features are arranged in not only spatial but also temporal directions like those in a video, it may be possible to compress them more efficiently by increasing a temporal correlation. To explore this possibility, we propose a “spatio-temporal arrangement”. This method spatially arranges the deep features as images and temporally arranges them as a video with a novel ordering search algorithm. Our method effectively increases the spatial and temporal correlations hidden in the deep features and achieves high compression efficiency compared with the previous methods. Experimental results demonstrate the compression efficiency of our method is better than that of the previous methods (1.50% to 4.98% on BD-Rate evaluation in a lossy setting). Our analysis shows that our method effectively increases the correlation when the input is an image with rich edges and textures.

1 January 2011
Verified
Conference Paper
Closed access
MIMasayuki Inoue
Masayuki Inoue
HKHideaki Kimata
Hideaki Kimata
KFKatsuhiko Fukazawa
Katsuhiko Fukazawa
NMNorihiko Matsuura
Norihiko Matsuura
This paper examines delivery methods for the interactive panoramic video delivery service that enables users to watch high-resolution panoramic images beyond HD quality using available network resources and the TV sets in users' homes. We propose a new data format based on Multiview Video Coding and a partial delivery method that permits even restricted bandwidth networks to support high-quality partial panoramic video transmission. We show the experiments conducted on the proposed partial delivery method. The results show that two types of the proposed method can achieve higher subjective video quality than the deliver-all method when the ROI (Region Of Interest) moves.

23 June 2023
Verified
Conference Paper
Open access
YIYoshiki Inazu
Yoshiki Inazu
HKHideaki Kimata
Hideaki Kimata
There have been many studies on image compression using deep learning methods. The most of these studies have focused mainly on increasing the image compression ratio. However, for effective implementation in a wide range of mobile devices, it is crucial to also develop lightweight network architectures that can cope with the limited computational resources and power constraints of these devices. We have proposed three image compression methods based on the MBConv (Mobile inverted Bottleneck Convolution) architecture, which is proposed for image recognition tasks in mobile devices. The main goal of this research is to improve the performance of image compression while at the same time reducing the computational complexity of the network. To validate the effectiveness of the proposed method, we constructed a network based on previous research and conducted a series of experiments. Then, using Rate-Distortion curves, we compared the performance of the proposed method with that of a comparison method. The results showed that all three proposed methods improve the peak signal-to-noise ratio (PSNR), an objective evaluation metric, especially at medium and low bit rates. In addition, the proposed methods were shown to successfully reduce the number of parameters in the network by up to 22%.


14 March 2024
Verified
Conference Paper
Open access
YIYoshiki Inazu
Yoshiki Inazu
HKHideaki Kimata
Hideaki Kimata
Extensive research on image compression has focused mainly on the 3-channel RGB format. However, many real-world applications, including image editing, icon and logo creation, frequently use the RGBA format, which incorporates an additional "alpha" channel to adjust transparency. In this paper, we propose a new deep learning-based network architecture specifically designed for RGBA images. The proposed architecture consists of an RGB encoder to process 3-channel input images, an alpha encoder dedicated to 1-channel input images, an RGB decoder to generate RGB images from latent variables, and an alpha decoder to generate alpha images. To compress image signals with the alpha channel, the proposed network architecture takes into account that the compressed image signal is not limited to a rectangular shape and is able to take any shape. Furthermore, a new loss function for training the network is proposed to account for the different signal characteristics between the RGB signal representing the image and the alpha channel depicting shape and transparency. In the evaluation, to train the network, the COCO dataset commonly used for semantic segmentation tasks was used as training image data and the labeled segmentation regions were used as alpha channels. The proposed method was compared with existing compression methods that support RGBA, such as BPG, WebP and AVIF. Experimental results show that the proposed method improves rate distortion performance across all bit rates.

19 September 2025 in IEEE Access
Verified
Journal Article
Open access
TSTakayuki Sasaki
Takayuki Sasaki
RTRyuichi Tanida
Ryuichi Tanida
MKMasaki Kitahara
Masaki Kitahara
YHYusuke Hiwasaki
Yusuke Hiwasaki
HKHideaki Kimata
Hideaki Kimata
YBYukihiro Bandoh
Yukihiro Bandoh
This paper proposes a fast-parallel method for singular value thresholding, aimed at low-rank analysis of many small matrices. In low-rank analysis, the problem of regularizing the nuclear norm, defined as the sum of singular values, is typically solved by iteratively applying singular value thresholding. However, this process often requires a long run-time due to the computational cost of singular value decomposition. To address this issue, we provide a method that is computationally inexpensive and can be executed in parallel. The core of our approach lies in the discovery of geometric properties of the nuclear norm, leading to a computational process that is fundamentally different from conventional methods. Our method is fast because it does not require singular value decomposition, which has been a bottleneck in terms of computational complexity and parallelism. We design the computation method for two-column matrices as the main proposal, and also demonstrate an optional method for approximately solving the three-column matrices problem. Through experiments, the proposed method calculated singular value thresholding up to 65.81 times faster with higher calculation accuracy than conventional methods.We also applied our method to real data for noise removal in 3D models and confirmed the improvement in computational speed.