Kuniaki Tabata

High-speed image scaling for integrated document management

Document information often includes image data (eg. handwritten annotations, signatures, maps, sketches and photos) as well as alphanumeric and other coded data. Thus, the capability of handling these image data plays an important role in reforming current office-work activities. This paper presents a high-speed scaling method to enlarge or reduce digital images defined as two-dimensional pixel arrays. Scale factors are assumed to be rational numbers R/r and S/s for the vertical and horizontal directions, respectively, where R, S, r, s denote positive integers. Under this assumption, coordinate determination for image scaling is performed by simple table look-up and shift operations. This method excludes time-consuming arithmetic operations for coordinate determination. Thus, high-speed scaling is realized. Experiments showed that the execution time per output pixel is about 150 ns when the shift clock frequency is 8 MHz. The proposed algorithm has been successfully applied to a multi-functional office workstation featuring integrated functions of data processing, document processing and image processing.

High-speed rotation of digital images by raster scanning and table-lookup operations

A high-speed processing method of rotating two-dimensional digital images by arbitrary angles is proposed. This method rotates images by combining two skew transformations of raster scanning along X and Y axes. It makes possible group access to horizontally or vertically adjacent pixels in the memory, and shortens memory read/write time per pixel. Each skew transformation includes scale change of images in which the scale factor is approximated by a rational number. The resultant periodicity in transformation is used for high-speed processing by simple table look-up operation. The estimated rotation time of images is 250 - 375 ns per pixel with a two-dimensional memory of transfer speed 1 ms/word (1 word = 16 bits) and the pixel shift clock time 125 ns (8 MHz). The scaling error due to rational number approximation depends on the transformation table capacity. The table capacity being 32 or 16 digits, the error is approximately 0.3 or 0.8 percent, respectively. In applying to document image editing, the table of 16 to 32 digits is practically sufficient, implemented by a small number of ICs (about 100). The effectiveness of this method is justified with those examples.

Matching of distorted central symmetry image—An application to seal impression collation

This paper considers the pattern matching of images in which, originally, the boundary has central symmetry which is made incomplete by defects and noises. A detection method for the centroid of symmetry is proposed. In the proposed method, the boundary is rotated by 180° to match itself (self-matching of the boundary), thereby detecting the centroid by utilizing all position information of the pixels on the boundary. When the defects are assumed, methods such as tracing cannot be applied directly to the extraction of the boundary, and a simple scheme is employed which scans the image in parallel to the coordinate axis. The effect of the spurious information produced by the scanning is eliminated by using the histogram as the measure of the boundary matching. The effect of the missing information is eliminated by superposing the histogram, which is obtained by scanning in the direction perpendicular to the original scanning, to the original histogram. The proposed method is applied to collation of seal impressions. It was verified that the positioning is performed within the error of one pixel for the ordinary defect of the impression frame and the partial defect of the frame due to the noise by background and fibrous padding material. It was also verified that a processing with practically sufficient speed can be realized. The proposed method will also be applied effectively to the positioning of the images with partial central symmetry, such as a circular component with a protrusion.