Perform a threshold segmentation for extracting characters.

The main application of char_threshold is to segment single-channel images of dark characters on bright paper. The operator works as follows: First, a histogram of the gray values in the image Image is computed for the points in the region HistoRegion{}. To eliminate noise, the histogram is smoothed with the given Sigma (Gaussian smoothing). In the histogram, the background (white paper) corresponds to a large peak at high gray values, while the characters form a small peak at low gray values. In contrast to the operator bin_threshold, which locates the minimum between the two peaks, here the threshold for the segmentation is determined in relation to the maximum of the histogram, i.e., the background, with the following condition:

       histogram[threshold] * 100.0 < histogram[maximum] * (100.0 - Percent)

For example, if you choose Percent = 95 the operator locates the gray value whose frequency is at most 5 percent of the maximum frequency. Because char_threshold assumes that the characters are darker than the background, the threshold is searched for ``to the left'' of the maximum.

In comparison to bin_threshold, this operator should be used if there is no clear minimum between the histogram peaks corresponding to the characters and the background, respectively, or if there is no peak corresponding to the characters at all. This may happen, e.g., if the image contains only few characters or in the case of a non-uniform illumination.

read_image (Image, 'letters')
char_threshold (Image, Image, Seg, 0.0, 5.0, Threshold)
connection (Seg, Connected)

char_threshold is reentrant and automatically parallelized (on tuple level).

anisotropic_diffusion, median_image, illuminate

connection, select_shape, select_gray

bin_threshold, auto_threshold, gray_histo, smooth_funct_1d_gauss, threshold

Foundation