Toshiharu Kurosawa

Dynamic range compression preserving local image contrast for digital video camera

This paper presents a novel dynamic range compression method for digital video cameras. The proposed algorithm is designed to preserve a local image contrast based on a luminance ratio of a pixel to its local surround in attention. In general, digital video cameras are required to maintain the shadow to middle range luminance that includes a main object such as a facial area. According to this requirement, most digital video cameras have adopted a knee curve as a dynamic range compression function that strongly compresses only the highlight range over the knee point. But this method has the weakness that severely lowers a highlight contrast. To overcome this weakness, the proposed algorithm automatically and adaptively enhances the local image contrast in the highlight regions. As a result, the highlight area such as blue sky is realistically reproduced without lowering the input luminance in the shadow to middle range.

Improvement of color quality with modified linear multi-scale retinex

With popular of Digital Still Camera-DSC, higher image quality is required. One of the subjects is that image quality at shadow area caused by the narrow dynamic range of the CCD devices is improved automatically. Conventionally, gamma transformation, histogram equalization, and etc. have been utilized for this improvement, but these are not always enough improvement. Recently, examinations applying to retinex theory taking into account of human eyes characteristics proposed by Land are paid attention. This algorithm renders image at shadow area clearly and effectively using spatial information between surrounding pixels arranged into two dimensions. Typical methods are Single-scale retinex(SSR) and Multi-Scale Retinex(MSR). These methods, however, does not always work on practical use in terms of color correction of the printed images with different RGB density distribution. In order to improve the issues of MSR, we propose the Modified Linear Multi-scale retinex (ML-MSR) method. A modified method consists of (a) linear computation processing and (b) synthesis both the original images and the images obtained by the linear MSR. By the simulation for the images printed by DSC, we show that ML-MSR can improve the visibility at shadow areas keeping with both the color balance and saturation, comparing with the conventional methods, such as histogram equalization and MSR proposed by Jobson. In general, a processing time of MSR remarkably increases with the size of Gaussian averaging filter to compute the weighted average. We describe about faster processing method of the ML-MSR algorithm, which has been shorten by using the thinning out of surrounding pixels and simplicity of average processing.

High dynamic range compression for digital video camera using local contrast enhancement

This work presents a high dynamic range compression technique that realistically reproduces the images captured by the digital video cameras. This technique adaptively enhances the local contrast of the highlight regions without lowering the luminance of the facial areas.

High quality color correction method combining neural networks with genetic algorithms

A variety of color correction technologies have been applied to hardcopies to eliminate the crosstalk caused by unwanted absorption in colorants. However, since the accuracy of color correction is not sufficient, an improvement in color correction technology is required as an aspect of the appearance of color devices. This paper proposes a new color correction method combining neural networks with genetic algorithms. An evaluation experiment of RMS color difference has been done with the chart pattern of 1331 colors inside the printer color gamut. As a result, it has been shown that the proposed method is qualitatively and quantitatively superior to the conventional color correction masking method using the least squares method.

Display. Special Contribution from Asia Display/IDW'01. Enhancement of JPEG Coded Images Using a Region-based Algorithm.

We describe an improved post-processing algorithm that reduces the blocking and ringing artifacts caused by quantization noise of the DCT coefficients in block-coding techniques such as JPEG without excessively blurring. It is a region-based algorithm that is iteratively applied based on projection onto convex sets. It first segments the decoded image into three regions : the region where blocking artifacts are strongly presented, the region consisting of edge pixels, and the region where ringing artifacts are presented. A process for reducing quantization noise suitable for each region is then calculated. Finally, the DCT coefficients are modified in accordance with the quantization constraint. Some experimental results demonstrate that this algorithm effectively removes the blocking and ringing artifacts in each region without excessively blurring.

A high-speed halftoning processor for raster scanned images

A high-speed bilevel reproduction algorithm, called modified error diffusion (MED) algorithm, has been developed to provide high-quality halftoning images for continuous tone images and has been implemented in a CMOS LSI chip. The chip has been designed with standard-cell 1.5- mu m CMOS technology using an optimum layout design. The chip has achieved a maximum processing speed of 60 ns/pixel. >