In this paper, image retrieval method using the features extracted in areas of subband division after the wavelet transform is discussed. Similar image can be retrieved in high accuracy by using the feature calculated by color values and edge positions in subband areas. The feature of color values is extracted by using the color histogram in low frequency area and the feature of edge positions is extracted by the higher order local autocorrelation. The retrieval simulation of similar image was performed using 500 various kinds of landscape images, and the good reference result whose reproduction rate is 90% was obtained. And the simulation shows that similar image reference was possible also in different picture resolution.
In medical field, Volume Rendering (VR) is useful in detecting cancer at an early stage. To achieve real-time VR of high-resolution volume data (VD), we propose a new parallel VR method for distributed-memory parallel machines. Our method is based on Segmented Ray-Casting and gives us (I1) lower communications at image compositing and (I2) load balance among processors. The method achieves (I1) by selecting processors at each compositing stage and (I2) by adjusting the size of sub-volume data on each processor. As a result, our method is about 2.7 times faster than conventional method on a PC cluster with 128 processors, so that achieves real-time VR of a 1024^3 VD at 1.5 frames/sec.
In this paper, the extraction method by human information of skin color, hair color, head shape, lip region and optical flow is proposed. Especially, we are able to determine whether occluded facial regions or only one facial region by using the distribution of optical flow. After this determination process, in the case that it was recognized as only one facial region, one facial region is extracted. Similarly, in the case that it was recognized as occluded facial regions, the priority order of facial region is decided and segmentation is started. Computer simulation shows the effectiveness for occluded facial regions, and extraction rate of facial regions 91.7% is obtained for 1929 frames.
This paper presents an extracting method of facial parts. In the extraction of faces that has already been presented in our laboratory, the facial parts are roughly extracted only for the evaluation whether an extracted area is a face or not. Accordingly, it was not necessary to extract facial parts precisely in this stage. However, the precise tracking of facial part's movement is required for the recognition of facial expressions. For this purpose, the facial parts must be precisely extracted. Computer simulation to extract facial areas using 37,416 frames shows more than 99.0% extracting accuracy, and which to extract mouth and eye areas shows 99.2% and 98.4% extracting accuracy, respectively. From the experimental results, the prospects of using this method are very encouraging.
We report current research issues about biometric person authentication. First, we introcuce any type of biometric feature for person authentication, and we point out problems in each biometric feature recognition technique. Next we point out difficulties in biometric person authentication technique regarded as image processing problems. Finaly, we report research databases for biometric person authentication.
In this paper, a robust personal identification system which is adaptable for facial images put on and taken off spectacles is presented. A human face identification system based on the isodensity maps has been already presented. It has been confirmed that the three dimensional personal features are well refrected in the isodensity maps. On the other hand, when the registered image taken off the spectacles and the input image put on it, the shapes of the input isodensity maps are extremely caused changes by the spectacles and are different from the registered isodensity maps. Then, the identification accuracy was the deteriorated. Therefore, parts of the spectacles are extracted from the input facial image, and the matching accuracy is improved.
In this paper, three-dimensional image of the face is acquired by the rangefinder, and the range image is generated from three-dimensional image. The estimation method of a face direction along horizontal and vertical axis from nose position is proposed. If the direction of the face can estimated automatically, it is possible to use as a human interface. Simulation results for three-dimensional image using other acquisition method shows that the validity of the proposed method is verified.
In this paper, a new effective method for isolating spectacles from facial images is presented. In the biometric system through facial images, many faces with spectacles must be taken into account. In general, the spectacles are appeared as a strip area in the image. By enhancing the strip area, the spectacles are then isolated.and removed. By interpolating the pixels, which are removed as the pixels included in the area of spectacles, the facial area after removing spectacles is obtained. The interpolation is executed with neighborhood pixels. Computer simulation using the 20 facial images shows a 74.3% correct extracting accuracy
To tracking of the movement of each specific facial area caused by the facial expression using video, it is thought required to compare a neutral face with an expressive face. From this point of view, we propose a facial expressions recognition system which is performed by iritegrative judgment of each movement of a specific facial area based on automatic detected neutral face. In this system, the recognition is performed by using not usual expert system, but the expert system in which the order of application of the rule set is set based on the characteristics of the expressions.