Skip to main content

Hitachi

Corporate InformationResearch & Development

Visibility Enhancement Technology for Cameras

— Presentation at IAPR MVA2013 —

June 27, 2013

Report from Presenter

The 13th IAPR International Conference on Machine Vision Application (MVA2013) was held from May 21 through 23, 2013 at Suzaku Campus, Ritsumeikan Univ., Kyoto, Japan. About 200 participants attended the conference, and the wide range of topics from basic theory to its application about machine vision was brought up for discussion. Yokohama Research Laboratory, Hitachi Ltd. introduced a poster presentation entitled "Visibility Enhancement Technology for Cameras" concerning contrast enhancement of image signals taken by cameras.

These days, industrial cameras represented by surveillance and vehicle-mounted cameras have become widespread and commonly used. The largest requirement for these cameras is to provide images in which every object is clearly recognizable. However, there are difficult shooting conditions such as high contrast scene (for example, shooting sunny outside with a camera in a dark room). It is an important issue to establish a low-cost visibility enhancement technology, which can be adapted to embedded equipments.


Fig. 1 Overview of the proposed
algorithm

Enlarge


Fig. 2 Example of the processed image
Enlarge

We proposed a signal processing algorithm which provides visibility enhancement with low amount of calculation by introducing the basic idea of illumination correction of the Retinex theory. The proposed algorithm adopts (1)tone redistribution to the input image depending on local luminance distribution, followed by (2)extraction and correction of illumination and reflectance, and (3)adaptive histogram optimization (see Fig. 1). At first, the input image is separated into some small regions, and luminance histograms are calculated for each region. The luminance level is adjusted so that the regional contrast is enhanced. Then, the processed image is separated into two components, which are called illumination and reflectance. The illumination component is adjusted to be equalized, while the reflectance component is adjusted to be enhanced. This process cancels the influence of illumination and yields an image in which the objects in dark areas and bright areas are both recognizable. Finally, the luminance value of each pixel is readjusted depending on the global luminance distribution.

The example of an image processed with the proposed method is shown in Fig. 2. It is confirmed that the signal level of the dark area is enhanced effectively while the tone of the bright area is maintained.

We also implemented the proposed algorithm in an FPGA-based system. We confirmed that the system is capable to perform video processing such as backlight correction for high definition movies in real-time.

(By YOSHIDA Daisuke)

  • Page top