Luminance Perception of the Human Visual System

With respect to quantization, it is important to know how the human visual system perceives the levels and what luminance diff erences can be distinguished. Figure 2.8 demonstrates that the small rectangle with a medium luminance appears brighter against the dark background than against the light one, though its absolute luminance is the same. This deception only disappears when the two areas become adjacent.

The human visual system shows rather a logarithmic than a linear re- sponse. This means that we perceive relative and not absolute luminance diff erences equally well. In a wide range of luminance values, we can re- solve relative diff erences of about 2%. This threshold value depends on

38                                                                                        2 Image Representation

 

A                                                                    b

Figure 2.9: A high-contrast scene captured by a CCD camera with a linear con- trast and a a small and b a large aperture.

 

a number of factors, especially the spatial frequency (wavelength) of the pattern used for the experiment. At a certain wavelength the luminance resolution is optimal.

The characteristics of the human visual system discussed above are quite diff erent from those of a machine vision system. Typically only 256 gray values are resolved. Thus a digitized image has much lower dynamics than the human visual system. This is the reason why the quality of a digitized image, especially of a scene with high luminance contrast, appears inferior to us compared to what we see directly. In a digital image taken from such a scene with a linear image sensor, either the bright parts are overexposed or the dark parts are underexposed. This is demonstrated by the high-contrast scene in Fig. 2.9.

Although the relative resolution is far better than 2% in the bright parts of the image, it is poor in the dark parts. At a gray value of 10, the luminance resolution is only 10%.

One solution for coping with large dynamics in scenes is used in video cameras, which generally convert the irradiance E not linearly, but with an exponential law into the gray value g:

g = Eγ.                                                       (2.9)

=         =

The exponent γ is denoted the gamma value. Typically, γ has a value of 0.4. With this exponential conversion, the logarithmic characteristic of the human visual system may be approximated. The contrast range is signifi cantly enhanced. If we presume a minimum relative luminance resolution of 10% and an 8 bit gray scale range, we get contrast ranges of 25 and 316 with γ 1 and γ 0.4, respectively.

For many scientifi c applications, however, it is essential that a linear relation is maintained between the radiance of the observed object and the gray value in the digital image. Thus the gamma value must be set to one for these applications.

2.3 Wave Number Space and Fourier Transform                               39

Figure 2.10: An image can be thought to be composed of basis images in which only one pixel is unequal to zero.

 

⇐ Предыдущая45678910111213Следующая ⇒

Поделиться: