We describe a statistical method to detect human behavior using appearance-based model of target objects. Most conventional gesture recognition systems utilize simpler method for the detection process such as background subtractions with assuming static observation conditions and those methods are not robust against camera motions, illumination changes and so on. In this paper, we propose a method to describe and recognize the appearances of target objects based on geometrical structures. Using the detection results, our system updates target tracking models. Human behavior is recognized by using the tracking models. Experimental results on human tracking and gesture recognition show the effectiveness of our method.
We propose a rapid face detection algorithm using Six-Segmented-Rectangle(SSR) Filter, combining range information and a template. First, we detect Between-the-Eyes as a face candidate point by scanning a certain size of rectangle which divided into six segments. Bright-dark relations of these six segments are tested if its center can be a candidate of Between-the-Eyes. Next, a true one among candidates is detected using range information and matching with a template. We implement the system on a PC with a xeon 2.2GHz CPU. This system processes image at 30 frames per second
We propose a method of human body detection and pose estimation from a complex image by using the simple feature extraction. We extract the elliptic shapes as head candidates and parallel lines as arms ans legs candidates from the image. Next the likelihoods of each body part are calculated for each candidates. The total likelihood of entire body is computed by combining them with knowledges about the connection of the positions between body parts. We construct the Parts Probabilities Connection Network, and optimize it by using the genetic algorithm. We report the experimental result to show the effectiveness of the proposed method.
Usually, 2-D wavelet transform consists of 1-D wavelet transform to the vertical and horizontal direction. Such 2-D wavelet is called ‘separable’ 2-D wavelet. Although we can analyze and synthesize local frequency using wavelet transform, we cannot distinguish between 45 degree component and 135 degree component using separable 2-D wavelet. The method proposed in this paper enables us to process an image locally and adaptively to various directions. Comparing results from image denoising experiments with the traditional method and the proposed method, we proved that the proposed method gives better results in respect of preservation of tilted edges.
A scalable bitstream syntax currently is used to realize the scalable video requirement. Compared with non-scalable video decoder, a much more complex video decoder is need to decode scalable bitstreams. However, it is considered that a video decoder should implement as simple as possible to play some quality video which users want. From this view point, a scalable video coding scheme by updatable bitstream is proposed to realize the scalability without scalable video decoder. In this paper, we pay attention to spatial updatable scalability and explain this algorithm. Finally, our proposed method is evaluated from the view point of the coding efficiency.
A commonly-accepted concept about the fixation method for joint prostheses is that a wide contact area between bone and artificial materials results in lower pressure that is better for biological tissues. Another concept is that the load should be applied to the proximal femur together with the distal femur because proximal and distal fitting can avoid stress shielding of bone. However, these two concepts appear to contradict each other because the exact contact point cannot be determined in an apparent and wide contact area. We postulated that limited fixation with a narrow contact area could be better for the initial fixation of a stem.
In this paper, we propose a technique for automatic scoliosis detection method from moiré topographic images. Normally the moiré stripes show a symmetric pattern as a human body is almost symmetric. According to the progress of the deformity of a spine, asymmetry becomes larger. Numerical representation of the degree of asymmetry is therefore useful in evaluating the deformity. First, displacement of local centroids is evaluated statistically between the left-hand side and the right-hand side regions of the moiré images with respect to the extracted middle line. The degree of the displacement with respect to normal and abnormal cases is learned by a neural network employing the BP algorithm. An experiment was performed employing 1,200 moiré images (600 normal, 600 abnormal) and 89% of the images were classified correctly by the NN.
As a standard motion estimation method, the template matching has been used for estimating relatively coarse motion field, wherein each estimated displacement represents the motion of the area with typically 8 by 8 or more pixels. This paper proposes a fast algorithm for the template matching method suitable for estimating high density or fine motion field, wherein each estimated displacement represents the motion of the area with 4 by 4 or less pixels. It is found that the computational complexity of the proposed algorithm is almost independent from the density of motion field and the size of templates.
In this paper, a new chromatic adaptation method called maximum color separation (MCS) is described. In the new method, the assumption is that image gamut reaches its maximum when under white illuminant. The essence of the MCS lies in the rigid theoretical framework of the coincidence of the chromatic gamut centroid with the coordinate of the ideal white illuminant when the area of a chromatic gamut is maximized. Inversely, the MCS explains the theoretical structure of the assumption. The MCS has strong similarity with the gray world, but essential differences in adaptation structures are shown. Experimental results using real images show the efficiency of the proposed method.
There have been many researches related to metamerism, because of the importance. Among them, Cohen R-Matrix systematized G. Wyszecki’s hypothesis in which a spectral radiance distribution is decomposed into the fundamental component and the metameric black component. This article discusses related to R-Matrix. We have derived the equation describing metameric blacks from our original point of view. First, it is proved that our equation is equivalent to the logic in the R-Matrix operation. Second, the zero-crossing problem of the metameric black is discussed, extracted from the spectral radiance distribution of equal energy by using R-Matrix.
It is important to recycle the old wood used in houses. Recently it is investigated to obtain charcoal from used wood. On the other hand, CCA treated wood is included in used wood. Because of including Arsenic, treated wood is not good for atmosphere in carbonization to obtain charcoal with high temperature. So we have tried to select out CCA treated wood from non-treated one. Surface of treated wood is a little bit greenish compared with non-treated one, so we first investigated the difference of the color phases. There are some differences for newly treated wood, but it is very difficult to find the method to select used CCA treated wood actually. We tried to detect CCA color from the surface of used wood using many kinds of chromaticity diagram.
JPEG compression proceeds in LSI’s inside of Special Machine as Digital Camera. Under this condition We cannot insert digital watermarks into this LSI’s inside. Then We report the result of digital watermark application as follows, “Before JPEG-compression Watermarks are embedded and after JPEG-decompression Watermark’s information are extracted”. We used Wavelet-transform, Convolution-coding, and Viterbi-Decoding effectively for it.
This paper describes a method for estimating the spatial distribution and the spectral distribution of omnidirectional illuminations in natural scenes, and an application of the method to image rendering in computer graphics. First, we introduce an imaging system using a spherical steel ball and a normal color CMOS camera. In order to get precise omnidirectional illumination distribution, we develop a mechanism for rotating the camera around the ball in a natural scene. We determine the mapping between the coordinates on the ball and the directional vectors of light rays in the world. Next, illuminant spectral distribution at each direction is estimated from RGB sensor data by using the linear finite model and the camera sensitivity functions. Moreover, the estimated spectral radiance distribution is used for image rendering of virtual objects. In experiments, an omnidirectional illumination is estimated in an indoor scene, and an CG image of metallic objects is created using the estimated illuminant.