Kazuo Ohzeki - Academia.edu (original) (raw)

Kazuo Ohzeki

Uploads

Papers by Kazuo Ohzeki

Research paper thumbnail of Video Analysis for Detecting Eye Blinking using a High-Speed Camera

2006 Fortieth Asilomar Conference on Signals, Systems and Computers, 2006

Without introducing pre-processing of eye region detection, the eye blinking motion was detected ... more Without introducing pre-processing of eye region detection, the eye blinking motion was detected from specific characteristic moving features on a face using a high-speed camera with a frame rate of 250 Hz/sec. In the process of closure and opening during eye blinking, a series of moving points has characteristic features. Both patterns of matching methods and many starting points on

Research paper thumbnail of Robust Watermarking Method By Systematic Block Diffusion Using Discrete Cosine Transform

Digital watermarks have long been considered as a security feature. A watermarking method that in... more Digital watermarks have long been considered as a security feature. A watermarking method that involves the diffusion of limited watermark information into a large part of an image’s data has high robustness. The diffused information is summed up to a single component before detecting the watermark. The summing up process eliminates small noises by an averaging effect, which improves the robustness of the embedded watermark against attack. In this field, thus far, only an asymmetrical Chirp transformation with a small block size has been attempted. In this study, a new verification experiment for a large block size of 256 × 256 pixels is conducted. High robustness of the proposed method is revealed. This includes the finding that, in the case of a JPEG compression attack, the proposed system is robust even at strong compression of 1/70. As for a clipping attack, embedded watermarks can be detected with up to seven pixel clipping of an embedded image.

Research paper thumbnail of A New Hough Transform Detecting Thick Line by Analysis of Vote Distribution -- Using Only Hough Space Without Refferring the Original Space

Research paper thumbnail of インタラクティブ知的符号化におけるグローバル動きパラメータの検出精度

The Journal of the Institute of Image Information and Television Engineers, 1998

Research paper thumbnail of Adaptive KL transform coding and its design method

Research paper thumbnail of Adaptive karhunen-loeve transform coding and its design method

Electronics and Communications in Japan (Part I: Communications), 1995

Specific models have been conventionally used to switch between various Karhunen-Wve (KL) transfo... more Specific models have been conventionally used to switch between various Karhunen-Wve (KL) transform matrices to adapt to local characteristics of nonstationary images. Here, the superior characteristic of the I U transform (mean square error minimization) is extended over all parts of the design process as an evaluation criterion. Further, by using actual images as training data, a design process is established that reflects the nature of real images. In order to compare the adaptive matrix-switching KL transform coding method with previous simple discrete cosine transform (DCT) coding methods, a transform is proposed that does not cause mean value separation. The effectiveness of this type of adaptive transform is determined by computer simulation. Also, since the resulting coding characteristics do not depend on the training images, it is estimated that the construction of a system that includes common nonstationary characteristics has been achieved. Examination of intermediate steps of this design process show that, in fact, all generated eigenvalues are practically positive and the stability of the design process is guaranteed.

Research paper thumbnail of Video Analysis for Detecting Eye Blinking using a High-Speed Camera

2006 Fortieth Asilomar Conference on Signals, Systems and Computers, 2006

Without introducing pre-processing of eye region detection, the eye blinking motion was detected ... more Without introducing pre-processing of eye region detection, the eye blinking motion was detected from specific characteristic moving features on a face using a high-speed camera with a frame rate of 250 Hz/sec. In the process of closure and opening during eye blinking, a series of moving points has characteristic features. Both patterns of matching methods and many starting points on

Research paper thumbnail of Robust Watermarking Method By Systematic Block Diffusion Using Discrete Cosine Transform

Digital watermarks have long been considered as a security feature. A watermarking method that in... more Digital watermarks have long been considered as a security feature. A watermarking method that involves the diffusion of limited watermark information into a large part of an image’s data has high robustness. The diffused information is summed up to a single component before detecting the watermark. The summing up process eliminates small noises by an averaging effect, which improves the robustness of the embedded watermark against attack. In this field, thus far, only an asymmetrical Chirp transformation with a small block size has been attempted. In this study, a new verification experiment for a large block size of 256 × 256 pixels is conducted. High robustness of the proposed method is revealed. This includes the finding that, in the case of a JPEG compression attack, the proposed system is robust even at strong compression of 1/70. As for a clipping attack, embedded watermarks can be detected with up to seven pixel clipping of an embedded image.

Research paper thumbnail of A New Hough Transform Detecting Thick Line by Analysis of Vote Distribution -- Using Only Hough Space Without Refferring the Original Space

Research paper thumbnail of インタラクティブ知的符号化におけるグローバル動きパラメータの検出精度

The Journal of the Institute of Image Information and Television Engineers, 1998

Research paper thumbnail of Adaptive KL transform coding and its design method

Research paper thumbnail of Adaptive karhunen-loeve transform coding and its design method

Electronics and Communications in Japan (Part I: Communications), 1995

Specific models have been conventionally used to switch between various Karhunen-Wve (KL) transfo... more Specific models have been conventionally used to switch between various Karhunen-Wve (KL) transform matrices to adapt to local characteristics of nonstationary images. Here, the superior characteristic of the I U transform (mean square error minimization) is extended over all parts of the design process as an evaluation criterion. Further, by using actual images as training data, a design process is established that reflects the nature of real images. In order to compare the adaptive matrix-switching KL transform coding method with previous simple discrete cosine transform (DCT) coding methods, a transform is proposed that does not cause mean value separation. The effectiveness of this type of adaptive transform is determined by computer simulation. Also, since the resulting coding characteristics do not depend on the training images, it is estimated that the construction of a system that includes common nonstationary characteristics has been achieved. Examination of intermediate steps of this design process show that, in fact, all generated eigenvalues are practically positive and the stability of the design process is guaranteed.

Log In