Image Processing Reference
3 Description of highlight image filter
This filter decreases more the values of pixel components (i.e., Red, Green, and Blue) that are
contrasting the image (which is similar to how Smart Contrast filter performs in the way that it
chooses between the same transformations as Smart Contrast) and from the other two proper-
ties that are sharpening and highlighting the edges of the features in the image (e.g., the char-
Highlight filter emphasized the areas of rapid intensity change (i.e., edges) like LoG (La-
placian of Gaussian) filter would do. Once the edges are detected, they are being sharpened,
which would produce a visual effect that is similar to what Sharpen filter would do to an im-
age. Still, the implementation of Highlight image filter has no similarities with Sharpen and
LoG filters' implementation.
In addition, Highlight filter creates shadows behind characters (the color of the shadows
contrasts with the color of the characters) and those shadows create a uniformly colored area
that covers the noise around characters and this way it leaves no isolated noise pixel around
the text that could damage the correctness of the OCR process. In a specific stage of it, the
OCR engine transforms the colored filtered text image into a black and white one in which the
contours of the characters are unbroken because of the uniformly colored shadows, turned to
whether black or white, that fill the characters (i.e., shadows will be white if the characters are
turned to white, and respectively, black if the characters are turned to black).
Precisely, all these combined properties of Highlight image filter, namely, selective contrast,
sharpening and highlighting the characters, and creating contrasting shadows behind text,
contribute to a high success rate of OCR on text images.
3.1 Description of the Image Filters' Visual Effects That Are
Included in Highlight's Visual Effect
Smart Contrast produces an effect similar to what Contrast does, that is increasing the contrast
of the image. In addition, Smart Contrast decreases more the values of pixel components that
are less than 127 and increases more the values greater than 127 than Contrast would do. High-
light filter contrasts the image in a way similar to Smart Contrast.
and could make the OCR produce worse results than with the unfiltered image. Highlight il-
ter takes the concept of spoting the edges from Sharpen ilter, but does not accentuate the
noise as well.
LoG combines the effects of Laplacian filter and Gaussian filter (which blurs the images in
order to reduce the sensitivity to noise). While Laplacian detects the regions of rapid intensity
change therefore being used in edge detection, LoG sharpens edges between two regions of
3.2 New Image Filter: Highlight
Highlight filter gathers together visual effects similar to the ones produced by Smart Contrast,
Sharpen, and LoG filters. The implementation is carried out in an original manner using no
template convolution (masks) like the last two mentioned filters do.