Page 1 - PerfectScan

PerfectScan White Paper – Copyright © Dynamic Computing Solutions 2013
1
PerfectScan ® – Beyond Digital Image Processing
and Enhancement
NOISE AND IMPERFECTIONS IN DIGITALLY RECORDED IMAGES
Lightning
Variation
Optical
Blurring
Quantification + Document Digital
image
Noise
Analog-to-Digital Conversion (Scanner, Digital Camera…)
Figure 1: Sources of noise and imperfection in a digitally recorded image
Noise is introduced into document images through either the conversion process or damages to the
physical documents itself. Noise can be viewed as coherent or incoherent with the underlying document
content.
Incoherent noise. Ink splotches, salt-and-pepper, stray marks, and marginal noise are representative of
the incoherent noise (with respect to the content). Since this type of noise shows a consistent behavior, it
is easier to detect and separate the noise from the content. Foxing, for example, can be classified as
incoherent as it has nothing to do with content. It can also be classified as coherent, as there is no known
systematic approach to separate it from the content.
Coherent noise. This type of noise occurs when the spatial frequency domain of the image cannot be
suppressed without prior knowledge of the content. Blur, pixel-shift and bleed-through are examples of
this type of noise. As the noise does not adhere to a consistent shape, position or size and tends to interact
with text in the foreground in irregular ways, it is comparatively more difficult to remove.
Image imperfection may be caused by several sources:
 Lighting variation. It is caused by uneven illumination and the shape and orientation of the page
relative to the light source.
 Optical Blurring. There are two main sources of blur. The first is lens imperfections or
aberrations. The second is displacement between the sensor array and the focal plane.
 Quantification. Light is integrated over small rectangles into pixels whose area determines the
level of quantization. Therefore, the adversary effect of quantification is inversely proportional to
the number of pixels in the sensor array, or the resolution of the captured image.
WHAT MAKES PERFECTSCAN® DIFFERENT?
There have been many attempts at perfecting analog-to-digital image enhancement. Traditional
techniques are applied at either the pixel or local level. These techniques span from the use of
morphological operators to reduce speckle, fill small holes and smooth edges, to the use of some forms of
dynamic thresholding. Although these techniques improve some images, they largely fail when there are
many different shading areas, uniform backgrounds or text on dark backgrounds.
PerfectScan® has been developed based on the early visual path processing and recognition models found
in the human visual system. Instead of processing an image at the pixel level, PerfectScan® takes it
further and analyzes the entire image using a set of cognitive-based algorithms. These neural-inspired