• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.98-101
• /
• 2019

At present, deep convolutional neural networks have made a very important contribution in single-image super-resolution. Through the learning of the neural networks, the features of input images are transformed and combined to establish a nonlinear mapping of low-resolution images to high-resolution images. Some previous methods are difficult to train and take up a lot of memory. In this paper, we proposed a simple and compact deep recursive residual network learning the features for single image super resolution. Global residual learning and local residual learning are used to reduce the problems of training deep neural networks. And the recursive structure controls the number of parameters to save memory. Experimental results show that the proposed method improved image qualities that occur in previous methods.

• Transactions of Materials Processing
• /
• v.16 no.4 s.94
• /
• pp.323-328
• /
• 2007

The residual stress in axial direction of the steel wires has been measured by using a method based on the combination of the focused ion beam(FIB) milling and digital image correlation(DIC) program. The residual stress is calculated from the measured displacement field before and after the introduction of a slot along the steel wires. The displacement is obtained by the digital correlation analysis of high-resolution scanning electron micrographs, while the slot is introduced by FIB milling with low energy beam. The experimental procedures are described and the feasibilities are demonstrated in steel wires fabricated with different conditions. It reveals that the tensile residual stress is formed in all steel wires and this is strongly influenced by the fabrication conditions.

• Journal of Information Processing Systems
• /
• v.18 no.1
• /
• pp.12-25
• /
• 2022

In recent years, deep convolutional neural networks have made significant progress in the research of single image super-resolution. However, it is difficult to be applied in practical computing terminals or embedded devices due to a large number of parameters and computational effort. To balance these problems, we propose CSRNet, a lightweight neural network based on channel split residual learning structure, to reconstruct highresolution images from low-resolution images. Lightweight refers to designing a neural network with fewer parameters and a simplified structure for lower memory consumption and faster inference speed. At the same time, it is ensured that the performance of recovering high-resolution images is not degraded. In CSRNet, we reduce the parameters and computation by channel split residual learning. Simultaneously, we propose a double-upsampling network structure to improve the performance of the lightweight super-resolution network and make it easy to train. Finally, we propose a new evaluation metric for the lightweight approaches named 100_FPS. Experiments show that our proposed CSRNet not only speeds up the inference of the neural network and reduces memory consumption, but also performs well on single image super-resolution.

• Journal of Information Processing Systems
• /
• v.18 no.3
• /
• pp.311-318
• /
• 2022

Tobacco sales enterprises often need to summarize and verify the daily sales bills, which may consume substantial manpower, and manual verification is prone to occasional errors. The use of artificial intelligence technology to realize the automatic identification and verification of such bills offers important practical significance. This study presents a novel multi-branch residual network for tobacco sales bills to improve the efficiency and accuracy of tobacco sales. First, geometric correction and edge alignment were performed on the input sales bill image. Second, the multi-branch residual network recognition model is established and trained using the preprocessed data. The comparative experimental results demonstrated that the correct recognition rate of the proposed method reached 98.84% on the China Tobacco Bill Image dataset, which is superior to that of most existing recognition methods.

• Journal of Information Processing Systems
• /
• v.19 no.3
• /
• pp.323-333
• /
• 2023

Facial expression recognition can aid in the development of fatigue driving detection, teaching quality evaluation, and other fields. In this study, a facial expression recognition method was proposed with a residual masking reconstruction network as its backbone to achieve more efficient expression recognition and classification. The residual layer was used to acquire and capture the information features of the input image, and the masking layer was used for the weight coefficients corresponding to different information features to achieve accurate and effective image analysis for images of different sizes. To further improve the performance of expression analysis, the loss function of the model is optimized from two aspects, feature dimension and data dimension, to enhance the accurate mapping relationship between facial features and emotional labels. The simulation results show that the ROC of the proposed method was maintained above 0.9995, which can accurately distinguish different expressions. The precision was 75.98%, indicating excellent performance of the facial expression recognition model.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.1
• /
• pp.63-71
• /
• 1993

An efficient RVQ image coding method is proposed using the segmentation-based coding and CVQ techniques. In the proposed method the residual image, the difference between an original image and the synthesized one obtained from the segmentation-based coding, is first dividel into $\times$4 subblocks. They are then individually coded in the spatial domain using a simple CVQ. Experimental results show that the proposed method yields better quality of the reconstructed images in both PSNR and subjective test over the basic VQ and SMVQ.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.1-15
• /
• 2022

The MobileNetV3 is specially designed for mobile devices with limited memory and computing power. To reduce the network parameters and improve the network inference speed, a new lightweight network is proposed based on MobileNetV3. Firstly, to reduce the computation of residual blocks, a partial residual structure is designed by dividing the input feature maps into two parts. The designed partial residual structure is used to replace the residual block in MobileNetV3. Secondly, a dual-path feature extraction structure is designed to further reduce the computation of MobileNetV3. Different convolution kernel sizes are used in the two paths to extract feature maps with different sizes. Besides, a transition layer is also designed for fusing features to reduce the influence of the new structure on accuracy. The CIFAR-100 dataset and Image Net dataset are used to test the performance of the proposed partial residual structure. The ResNet based on the proposed partial residual structure has smaller parameters and FLOPs than the original ResNet. The performance of improved MobileNetV3 is tested on CIFAR-10, CIFAR-100 and ImageNet image classification task dataset. Comparing MobileNetV3, GhostNet and MobileNetV2, the improved MobileNetV3 has smaller parameters and FLOPs. Besides, the improved MobileNetV3 is also tested on CPU and Raspberry Pi. It is faster than other networks

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.20 no.2
• /
• pp.375-384
• /
• 2016

In this paper, when stereo image, satellite view and corresponding depth maps were used as the input data, we propose a new method that convert these data to data format suitable for compressing, and then by using these format, intermediate view is synthesized. In the transmitter depth maps are merged to a global depth map and satellite view are converted to residual image corresponding hole region as out of frame area and occlusion region. And these images subsampled to reduce a mount of data and stereo image of main view are encoded by HEVC codec and transmitted. In the receiver intermediate views between stereo image and between stereo image and bit-rate are synthesized using decoded global depth map, residual images and stereo image. Through experiments, we confirm good quality of intermediate views synthesized by proposed format subjectively and objectively in comparison to intermediate views synthesized by MVD format versus total bit-rate.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.3
• /
• pp.96-108
• /
• 1996

In this paper, we propose an object-oriented coding method in low bit-rate channels using block-based motion vectors and residual image compensation. First, we use a 2-stage algorithm for estimating motion parameters. In the first stage, coarse motion parameters are estimated by fitting block-based motion vectors and in the second stage, the estimated motion parametes are refined by the gradient method using an image reconstructed by motion vectors detected in the first stage. Local error of a 6-parameter model is compensted by blockwise motion parameter correction using residual image. Finally, model failure (MF) region is reconstructed by a fractal mapping method. Computer simulation resutls show that the proposed method gives better performance than the conventional ones in terms of th epeak signal to noise ratio (PSNR) and compression ratio (CR).

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.21 no.12
• /
• pp.3033-3045
• /
• 1996

In this paper, we propse an object-oriented coding method in low bit-rate channels using pyramid structure and residual image compensation. In the motion estimation step, global motion is estimated using a set of multiresolution images constructed in a pyramid structure. We split an input image into two regions based on the gradient value. Regions with larte motions obtain observation points at low resolution level to guarantee robustness to noise and to satisfy a motion constraint equation whereas regions with local motions such as eye, and lips get observation points at the original resolution level. Local motion variations and intesity variations of an image reconstructed by the golbal motion are compensated additionally by using the previous residual image component. Finally, the model failure (MF) region is compensated by the pyramid mapping of the previous displaced frame difference (DFD). Computer simulation results show that the proposed method gives better performance that the convnetional one in terms of the peak signal to noise ratio (PSNR), compression ratio (CR), and computational complexity.