This article presents state of the art and future possibility of medical images. Since most of these images are digitally stored in computers, we can use them not only for diagnosis but also medical treatment planning such as surgical simulation. Medical image database such as PACS or ISAC have also important rolse in future medical image system.
Narrow-MUSE is a system which NHK has developed for a simulcast ATV system in the United States. Adopting the same multiple subsampling technique as used in the MUSE system, and the new modulatin scheme, the Narrow-MUSE can deliver a high-resolution picture in the 6-MHz bandwidth without causing intereference with existing NTSC services. ATV laboratory tests conducted by the ATTC and ATEL were completed in February 1992. The good performances of Narrow-MUSE in interferernce, and picture/audio quality were confirmed by the results of the tests.
This paper describes an experimental image compression/decompression system using Digital Image Signal Processor(DISP) VLSIs. The system which is constructed in a PC environment, consists of compression/decompression unit, digital storage unit and video input/output unit. A hybrid architecture is employed in the compression/decompression unit. This architecture combines software implementation by DISPs with hardware implementation by DCT-LSI. In order to distribute the computation loads for each DISP and to achieve high parallelism in the multi processor system, adaptive-function-assignment scheme has been proposed. Motion video decompression according to MPEG1 algorithm has been realized at the rate of 30fr/s for 176×l44pixel. JPEG algorithm, which is for still picture compression/ decompression, also has been implemented by the software based processing using DISP. In addition, image processings including special video effects have been realized.
The block matching technique, one of the methods for motion estimation of picture sequences, are widely used for motion-compensated picture coding and frame interpolation in TV standard conversion. As the criterion of matching, Minimum Mean Square Error (MMSE) and Minimum Mean Absolute Error (MMAE) are often used. However, it is not guaranteed that they give the visually true motion vector, especially when brightness level of the picture changes or different kinds of movement exist in a block, i.e., when the assumtion that the motion within a block is homogeneous does not hold, on which the block matching technique is based. For the purpose of obtaining the proper motion vector even when the above assumption does not hold, a new filtering scheme is presented. In this scheme, the input pictures to the block matching are filtered by a filter based on a human vision model, before entering the block matching unit.
A new image magnification method with extrapolation of spatial high frequencies is proposed. This method improves the image quality by restoring spatial high frequencies vanished in the observation. In this method, spatial high frequencies are restored by using a priori known constraints, as space-limitedness and band-limitedness, during the iterative transform with DCT for an image. Simulation results show that restoration errors are reduced and the image quality is improved as compared to three conventional interpolation methods.
A method for estimating reflectance parameters for 3-D objcect surface is described. A 3-D object is set on a turntable and is observed by a TV color camera. The object shape is measured using a range finder and the surface is approximated with polygonal patches. For each patch, reflected intensities for object surface are observed from multiple view points. The Cook-Torrance reflectance model is used and the reflectance parameters are estimated by using the least square estimation. To reduce the computation cost, the surface patches are classified baced on color infomation, and the reflectance parameters are estimated for each region. Computer generated images are obtained using the shape infomation and the reflectance parameters of the 3-D object surface.
Human face recognition (HFR) is important in many applications, such as access control and criminal identification. The recognition process in brain involves highly complex human psychological processing. So, the mechanism is still unknown. For constructing a practical system of HFR, it is necessary to develop the algorithm of stable and accurate face features extraction and the usable compound recognition process. We had developed a method for feature extraction of face image using multi-resolution image processing and Hough transform. In this paper, a method of human face identification using this algorithm and pattern matching is discribed and a face image generation system is proposed.