In this paper, a VLSI-oriented, fast reconstruction algorithm from projection data of a few directions is proposed, which is based on wavelets sampling model, singular value decomposition, part area reconstruction and uniformity scanning method. At first, by imbedding the classical Shannon sampling theorem in wavelet theory, reconstruction algorithm with wavelet sampling function has better efficiency and less error. Then, by means of combination of multiple wavelets sampling model and singular value decomposition, moderate CT image with least squares of error can be reconstructed by sum of products operation. In addition, as this algorithm is suitable for being achieved by massive parallel processing, it will be possible to develop faster, more compact and economic industry-oriented CT system. At last, The results of computer simulation with these reconstruction algorithms are presented.

View full abstract

Recently, many studies have been carried out in dealing with the automatic input of maps to a computer. The automation of a map processing makes up the labor storage. The purpose of this study is to extract roads from the urban map and restore the roads, automatically. In this paper, the automatic extraction method of roads from a map is proposed. In the case that roads in the map are hidden by symbols or characters, the restoration of roads are also carried out. In the extraction, center lines are obtained based on the property that roads are usually consist of parallel lines. The method detects a road network based on the parallel lines. Next, the reconstruction of roads is made using the center lines. Finally, computer simulations are made for confirmation of an algorithm. Using the many practical maps, the usefulness of the method is verified. There are some merits in processing time and practicality by the applications of labeling and noise removal.

View full abstract

In the surveillance of moving objects with TV cameras, snowfall prevents the clear surveillance. Therefore it would be desirable that snowflakes are eliminated from the surveillance images. In this paper, we tried to eliminate snowflakes by using time filtering. First we described time filter, then we estimated the performance of the filter theoretically by using a stochastic model that utilizes the size and velocity of a snowflake. Moreover, we considered the effect of the filter to the moving objects such as walking persons or running cars. Finally we applied the filter to the real images and obtained good results.

View full abstract

Many works have so far been made on digital watermarking, and it has been developed to the level of practical use. However all methods of watermarking proposed so far have no tolerance to attacks of local distortions. In this paper we propose a hierarchical method of watermarking which tolerates local distortions. Using this method, embedded watermark can be extracted against a considerable degree of local distortions, since the information is determined at a higher level than the original image. By this method, moreover, the degradation of image quality can be reduced due to the local determination of watermark strength based on the standard deviation of neighboring pixel values. Experimental results show that this method enables correct embedded information to be detected against local distortion attacks.

View full abstract

In this paper, the authors propose a method of identifying individuals using the lip shape as password. Identifying individuals are able to be carried out by recognizing the password and the characteristic shape of the lip from the facial image. First, among the sequence of the images, each object image for processing is detected by using mouth movement information. Next step, in each image, the mouth region is detected by the color information. Then, the three shapes which are outside line of lip, the inside line of lip, and the center line of lip are extracted by using the change in distribution of gray level. Finally, the recognition of phoneme and individuals are carried out by the method based on the subspace method. As the training, the subspace method needs the 20 features extracted from the above three shapes. The experiments using the experimental system are performed to demonstrate the efficacy of this method and the experimental results are shown.

View full abstract

It is an important problem to generate more natural and realistic CG images by CG simulation. In the case of the CG display of botanical tree images, the modeling of real and natural tree barks are a very important component to express the characteristics. In this study, to generate 3D shape of barks that have a wide range of variations and are much complicated, we propose a new method of generating bark images of various botanical trees by using characteristic parameters. In simulations based on this method, it turned out that we can generate realistic botanical bark images with the reduction of the amount of data to be saved and computational time.

View full abstract

This paper presents a Cellular Neural Network (CNN)-based algorithm to segment a texture image by combining some texture segmentation results. Due to the diversity of texture, using multiple segmentation results segmented by different algorithms is necessary for texture image segmentation problems. In this paper, a new method called Composition-Combination is proposed to combine some initial segmentation results. A new kind of CNN called Multi-objective CNN (MOCNN) is developed to improve the combination result of Composition-Combination and produce final segmentation. Different from the standard CNN, each cell of MOCNN has multiple vectors denote different features of cell, and one vector will occupy the cell against other vectors when the network gets to the equilibrium state.

View full abstract

Ridge lines are useful geometric feature for image analysis, in which mountain ridge lines are important in obtaining Digital Elevation Models (DEMs). This paper presents a wavelet-based method to detect mountain ridge lines for synthetic aperture radar (SAR) images. First, ridge lines of mountains as high frequency signals are emphasized by wavelet transform; then average variance judgement and histogram analysis, which are supplement of wavelet transform, are used to extract temporal mountain areas (MA) and final MA respectively; finally, Laplacian of Gaussian (LoG) filter is used to detect ridge lines in the final MA. The method is illustrated by 550×550 ERS-1/SAR data and satisfactory results are presented.

View full abstract