After entering the 21st century, mobile internet is advanced in volume and characteristics. Its services have expanded to include image information, other multimedia and social networks. In this paper, aiming to prepare for mobile internet era, the evolution of DRM technology will be surveyed. By looking back the demands for software protections, standardization of cypher algorithms and expansion of digital content distribution, I will review the process to commercialization of DRM technology, which started from conceptual proposal and achieved broad acceptance in the 21st century. Then I will suggest the importance of sustainability of the services and propose a new security model based on game theory.

View full abstract

Our ultimate goal of research is to develop a method that allows people to visualize the quality of their sleep without any expertise or extra care on health, which we hope achieves primary prevention of sleep-related diseases in daily life. Quality of sleep is known to be related to the cycle of two different kinds of sleep stages that are characterized by rapid eye movement. We propose a system that is capable of detecting the timing of rapid eye movement only by videotaping the face while sleeping. The system captures face using near infrared camera, detects a face with the eyes and the eyebrows in the initial video frame, fetches sub pixel values along the line that runs horizontally across the upper eyelid and generates an image intensity vector, and accumulates the vector over time to create an image that contains texture pattern correspondent to rapid eye movement. Experimental results demonstrate that the proposed method can successfully detect rapid eye movement for both simulated and real sleep with eye closure.

View full abstract

There is a growing need for content retrieval techniques along with the popularization of video content services through the network. For the coded video content, there is a problem in search speed, because previous methods require the decoding process to reconstruct the spatial information. Moreover, previous schemes using motion couldn't use the local moving data. So, I propose a video retrieval method based on motion information in this paper. It uses moving region extraction scheme by the local maximum in the histogram of motion vectors. As a motion feature information, the number of motion regions, the region area, the amount of movements, their coordinates were used. The method(a) using all motion vectors in a frame was defined. In addition, the method(b) used the feature values of the histogram of motion vectors. By comparing proposed method(c) with (a),(b) from the aspect of search precision and processing time, I considered the effect. In the simulation experiments, (c) improved performance of the same sequence identification in (b) by the local motion. Besides, in the whole system, (c) were about 27.64 and 1.37 times faster than (a) and (b), respectively. In the matching processing, it is also about 462 and 3.15 times faster in respective schemes. Accordingly, I could confirm the effectiveness of proposed method. As a further study, I will consider a video search method using the global and the local motion in a frame.

View full abstract

I have already proposed ANNG (Approximate k-Nearest Neighbor Graph) as a graphstructured access method to search for image features. However when ANNG is applied to real product image searches, two issues come up. First, when a large amount of data is stored into ANNG, some nodes on the graph are likely to have too many edges due to the characteristics of the ANNG method. Second, the existing combination search methods of image feature search and product category search are not enough to provide appropriate search results to users. To address the first issue, I propose a method to eliminate the excess edges while maintaining the connectivity of the graph. To address the second issue, I also propose a method of search using a single metric space which is a combination of both category tree-based category feature space and image feature space.

View full abstract

In recent years, research on recognition and modeling of objects to take advantage of sequential point clouds that was acquired using a range image sensor on move is active. In registration for integrating point clouds, there are correspondences based methods that has low accuracy but requires small computation time and ICP(Iterative Closest Point) algorithm that has high accuracy but huge computation time. Correspondences based methods depend on accuracy of feature points location and feature point matching because they require correspondences of point clouds using feature point matching. On the other hands, the computation time of ICP algorithm multiplies accordingly when the interval between point clouds is large. In this paper, we propose novel registration methods with 3D motion vector estimation of sequential point clouds. According to experiments, we confirmed that proposed methods are able to process with less computation time with equivalent accuracy than previous methods.

View full abstract

Smartphone users browse and download confidential document files routinely. Therefore, the higher security level is needed for smartphones than conventional mobile phones. For Android smartphones, there exist three authentication methods; the PIN authentication method, the password authentication method, and the pattern authentication method. In the pattern authentication method, to release the lock of smartphones users trace four or more of nine points appeared on the screen in the predetermined order. Therefore, the pattern authentication method can be considered as taking advantage of characteristics of a touch screen. However, there exists an attack for it called the smudge attack that is a way of attacking that attackers guess the authentication pattern from the smudge on the screen. Then, in this study, we propose a user authentication method for smartphones having the tolerance to smudge attacks. In the proposed method, users can release the lock of smartphones by selecting the pass-image from a number of images and flicking it. The method has also the tolerance to observation attacks and so on because the latest image shot by the user oneself is used as the pass-image, that is, the pass-image is updated at each time of taking pictures.

View full abstract

In recent years, opportunities of taking photographs have been increasing owing to the popularization of digital cameras, smart phones, and SNSs that we can post photos easily. Because of the convenience of taking photos using digital cameras, we often take many photos of the same object. However, it is difficult to select the best photo when the number of photos is large. Therefore, in this study, we propose a method to select the best photo from a group of photos of the same object. In the proposed method, first, a candidate region of the photographic subject is extracted by image segmentation based on color reduction processing. Next, the sharpness of the photographic subject region is calculated to each photo in the group, and the best photo is selected by comparing the values of the sharpness. The results of evaluation experiments conducted on multiple groups of photos of the same object show that the proposed method is useful to select the best photo.

View full abstract

It is important to generate realistic eye movements by Computer Graphics. However, previous methods require a lot of expense and work when we generate eye movements in conversations. In this paper, we propose a method which automatically generates eye movements during conversation based on probability models derived from actual measurements. First, we measure eye movements with blinks in conversations and fixation eye movements. Second, we classify eye movements into saccades and fixation eye movements based on measurement result. Third, we approximate saccades, fixation eye movements and blinks using probability models and apply them to characters. As a result, we can generate realistic eye movements automatically.

View full abstract

Film comic is created from animation movies by selecting important frames, trimming the frame images to fit into the panels of comic and placing speech balloons at appropriate positions on the panels. Conventional technologies aiming at automatically completing these tasks based on low level image features only fail to produce good results, because the detection of important frames and important areas in each frame requires the understanding of movie and image contents as well. We propose an iMap(informative Map) generating method based on viewers’ eye-tracking data and the feature of image for automatic creation of film comic.

View full abstract

Human tracking in monocular image sequences has been studied in the field of computer vision for many kinds of applications such as surveillance system, intelligent room, sports video analysis and so on. Human tracking in real environment is including issues caused by various factors. These include illumination variation, partial or complete occlusion of human body, and wide variety of body shapes. In this paper, we present a novel method for robust human tracking using statistical human model of appearance variation with postural change. This statistical human model can represent different postural appearances of human, and enables effective and robust human tracking with simple features such as silhouette, edge and color. Our proposed method realized more robust human tracking to not only partial occlusion but also appearance variation by likelihood evaluation of each body parts based on statistical human model. The experimental results validate the robustness of the proposed method to partial occlusion and human posture varieties.

View full abstract

In today’s traffic conditions with a lot of information, it is very important to get suitable information on the conditions. Among them, the detection of only traffic convex mirrors contributes greatly in terms of providing safety and very useful information for drivers. This paper emphasizes pre-processing and limits the candidate to target edge pixels in advance. First, specific processing content is noise reduction using curvature after the edge extraction. After that, removing straight line and detecting ellipse using Least Square Method or Hough Transform. In addition, specifying elliptical region of traffic convex mirror using the entropy after ellipse detection. The detection performance of Least Square Method is inferior compared with HT, however, the computational cost is small. On the other hand, the detection performance of Hough Transform is robust, while requiring a large computational cost. By comparing both methods, we validate the methods using pre-processing in this paper and confirm the effectiveness and versatility by the experimental results.

View full abstract

The reconstructed image quality can be made excellent by applying the error diffusion method for the computer-generated hologram because the object and the noise can be separated in the process. It is known that image quality is further improved when the error-diffusion method with edge emphasis, which changes the threshold pixel by pixel, is applied. However, there was no investigation concerning the optimization of both of the error diffusion coefficients and the edge emphasis coefficients so far. In this paper, we will introduce the new algorithm in which the error diffusion coefficients and the edge emphasis coefficients are designed at the same time by the genetic algorithm, and we will show that the optimization of the reconstructed image quality can be achieved through this method.

View full abstract