亀田 裕介

An autoregressive image generative model that estimates the conditional probability distributions of image signals pel-by-pel is a promising tool for lossless image coding. In this paper, a generative model based on a convolutional neural network (CNN) was combined with a locally trained adaptive predictor to improve its accuracy. Furthermore, sets of parameters that adjust the estimated probability distribution were numerically optimized for each image to minimize the resulting coding rate. Simulation results indicate that the proposed method improves the coding efficiency obtained by the CNN-based model for most of the tested images.

The power-generation capacity of grid-connected photovoltaic (PV) power systems is increasing. As output power forecasting is required by electricity market participants and utility operators for the stable operation of power systems, several methods have been proposed using physical and statistical approaches for various time ranges. A short-term (30 min ahead) forecasting method had been proposed previously for multiple PV systems using motion estimation. This method forecasts the short time ahead PV power generation by estimating the motion between two geographical images of the distributed PV power systems. In this method, the parameter λ, which relates the smoothness of the resulting motion vector field and affects the accuracy of the forecasting, is important. This study focuses on the parameter λ and evaluates the effect of changing this parameter on forecasting accuracy. In the periods with drastic power output changes, the forecasting was conducted on 101 PV systems. The results indicate that the absolute mean error of the proposed method with the best parameter is 10.3%, whereas that of the persistence forecasting method is 23.7%. Therefore, the proposed method is effective in forecasting periods when PV output changes drastically within a short time interval.

We propose a HDR (high dynamic range) reconstruction method in an image sensor with a pixel-parallel ADC(analog-to-digital converter) for non-destructively reading out the intermediate exposure image. We report the circuit design for such an image sensor and the evaluation of the basic HDR reconstruction method.

This paper proposes a lossless coding method for HDR color images stored in a floating point format called Radiance RGBE. In this method, three mantissa and a common exponent parts, each of which is represented in 8-bit depth, are encoded using the block-adaptive prediction technique with some modifications considering the data structure.

We previously proposed a novel lossless coding method that utilizes example search and adaptive prediction within a framework of probability model optimization for monochrome video. In this paper, we improve the adaptive prediction in terms of coding performance and processing time. More precisely, we made modifications to the following three items: (a) reference pel arrangements, (b) motion vector derivation, and (c) optimal selection of predictors. Experimental results show that the proposed method certainly improves the coding performance and the processing time compared to our previous method, and achieves better coding performance than the VVC-based lossless video coding scheme.

This paper reports the evolution and application potential of image sensors with high-speed brightness gradient sensors. We propose an adaptive exposure time control method using the apparent motion estimated by this sensor, and evaluate results for the change in illuminance and global / local motion.

We propose an adaptive exposure-time-control method for image sensors, which can control the exposure time for each pixel to reconstruct a high-dynamic-range image, while suppressing blown-out highlights and blocked-up shadows, according to the luminance and contrast of the scene. First, the proposed method determines the exposure time that maximizes the entropy of the entire image, as an image with high entropy contains more object details. In order to estimate the exposure time appropriate for the light and dark areas in the scene, the proposed method divides the image into blocks and estimates the exposure time that maximizes the entropy for each block. Because the proposed method captures and estimates several exposure times simultaneously, the time required for adjusting the exposure time is reduced. Simulation experiments show the effectiveness of the proposed method.

Photon counting imaging can be used to capture clearly photon-limited scenes. In photon counting imaging, information on incident photons is obtained as binary frames (bit-plane frames), which are transformed into a multi-bit image in the reconstruction process. In this process, it is necessary to apply a deblurring method to enable the capture of dynamic scenes without motion blur. In this article, a deblurring method for the high-quality bit-plane frame reconstruction of dynamic scenes is proposed. The proposed method involves the deblurring of units of object motion within a scene through the application of motion compensation to pixels sharing the same motions. This method achieves more efficient motion blur suppression than the application of simple deblurring to pixel block or spatial region units. It also applies a novel technique for accurate motion estimation from the bit-plane frame even in photon-limited situations through the statistical evaluation of the temporal variation of photon incidence. In addition to deblurring, our experimental results also revealed that the proposed method can be applied for denoising, which improves the peak signal-to-noise ratio by 1.2 dB. In summary, the proposed method for bit-plane reconstruction achieves high quality imaging even in photon-limited dynamic scenes.

Scene flow is a three-dimensional (3D) vector field with velocity in the depth direction and optical flow that represents the apparent motion, which can be estimated from RGB-D videos. Scene flow can be used to estimate the 3D motion of objects with a camera; thus, it is used for obstacle detection and self-localization. It can potentially be applied to inter prediction in 3D video coding. The scene-flow estimation method based on the variational method requires numerical computations of nonlinear equations that control the regularization strength to prevent excessive smoothing due to scene-flow regularization. Because numerical stability depends on multi-channel images and computational parameters such as regularization weights, it is difficult to determine appropriate parameters that satisfy the stability requirements. Therefore, we propose a numerical computation method to derive a numerical stability condition that does not depend on the color of the image or the weight of the regularization term. This simplifies the traditional method and facilitates the setting up of various regularization weight functions. Finally, we evaluate the performance of the proposed method.

In general, »drawing collapse» is a word used when very low quality animated contents are broadcast. For example, perspective of the scene is unnaturally distorted and/or sizes of people and buildings are abnormally unbalanced. In our research, possibility of automatic discrimination of drawing collapse is explored for the purpose of reducing a workload for content check typically done by the animation director. In this paper, we focus only on faces of animated characters as a preliminary task, and distances as well as angles between several feature points on facial parts are used as input data. By training a support vector machine (SVM) using the input data extracted from both positive and negative example images, about 90% of discrimination accuracy is obtained when the same character is tested.

This paper describes a method for creating cel-style CG animations of waving hair. In this method, gatherings of air are considered as virtual circles moving at a constant velocity, and hair bundles are modeled as elastic bodies. Deformation of the hair bundles is then calculated by simulating collision events between the virtual circles and the hair bundles. Since the method is based on the animator's technique used in creation of the traditional cel animations, it is expected to suppress a feeling of strangeness that is often introduced by the conventional procedural animation techniques.

Abstract This paper proposes a lossy image coding method for still images. In this method, recursive and non-recursive type intra prediction techniques are adaptively selected on a block-by-block basis. The recursive-type intra prediction technique applies a linear predictor to each pel within a prediction block in a recursive manner, and thus typically produces smooth image values. In this paper, the non-recursive type intra prediction technique is extended from the angular prediction technique adopted in the H.265/HEVC video coding standard to enable interpolative prediction to the maximum possible extent. The experimental results indicate that the proposed method achieves better coding performance than the conventional method that only uses the recursive-type prediction technique.

We previously proposed a lossless image coding scheme using example-based probability modeling, wherein the probability density function of image signals was dynamically modeled pel-by-pel. To appropriately estimate the peak positions of the probability model, several examples, i.e., sets of pels whose neighborhoods are similar to the local texture of the target pel to be encoded, were collected from the already encoded causal area via template matching. This scheme primarily makes use of non-local information in image signals. In this study, we introduce a prediction technique into the probability modeling to offer a better trade-off between the local and non-local information in the image signals.

We propose a method of interframe prediction in depth map coding that uses pixel-wise 3D motion estimated from encoded textures and depth maps. By using the 3D motion, an approximation of the depth map frame to be encoded is generated and used as a reference frame of block-wise motion compensation.