Automatic retrieval process for machine-printed or handwritten text on a background, in a multilevel digital image
OwnerJeffrey Horace Johnson, Jean-Claude Andre Simoninventors
Life of this patent
Estimating the term…
Abstract
Systems and methods for frequency acquisition and channel tracking in a direct sequence code division multiple access system are described. Frequency estimates are prepared and refined, which estimates are also used in the channel tracking process.
Claims10
A process for the automatic retrieval of machine-printed or handwritten text over a background, in a digital image with N grey levels, where N is an integer greater that 2, in accordance with which, for each group constituted by a line or several contiguous lines of pixels of the image:
- (a) the greyscale value of each pixel is compared with the respective values of at least two other pixels belonging to a predetermined neighbourhood of this pixel, and the values of the pixels which, after comparison, satisfy simultaneously at least two predetermined inequalities are retained;
- then, for each group of inequalities to be verified simultaneously:
- (b) a histogram is established giving, for each greyscale value between 0 and N-1, the number of pixels accepted at the conclusion of the comparisons carried out in step (a);
- (c) from the histogram is determined, among the N greyscale values, a threshold S;
- (d) the pixels accepted are divided into a first category, for which the value is above S, and a second category, for which the value is below S;
- (e) the said group constituted b one or more lines of the image is scanned horizontally, and vertically respectively, and on each line, and each column respectively, are determined horizontal, and vertical respectively, runs, as being groups of contiguous pixels horizontally, and vertically respectively, and which have been classified in step (d0 in the second category;
- (f) among the previously determined horizontal, and vertical respectively, runs, vertical polygons are read, including a number y.sub.v of vertically connected horizontal runs, and horizontal polygons, including a number x.sub.h of horizontally connected vertical runs;
- (g) when the total number y.sub.v of vertically connected horizontal runs is lower than a predetermined number p.sub.v or when the total number x.sub.h of horizontally connected vertical runs is lower than a predetermined number p.sub.h, steps (d) to (f) are repeated with a modified threshold value S;
- the said machine-printed or handwritten text being obtained in the form of the said retrieved polygons.
A process for the automatic retrieval of machine-printed or handwritten text over a background, in a digital image with N grey levels, where N is an integer greater that 2, in accordance with which, for each group constituted by a line or several contiguous lines of pixels of the image:
- (a1) the greyscale value of each pixel is compared with the respective values of at least two other pixels, and the values of the pixels which, after comparison, verify simultaneously at least two predetermined inequalities are accepted;
- (b1) the said group constituted by one or more lines of the image is scanned horizontally, and vertically respectively and on each line, and each column respectively, are determined horizontal, and vertical respectively, gradient runs, as being groups of contiguous pixels horizontally, and vertically respectively, and which have been accepted at the conclusion of the comparisons carried out in step (a1);
- (c1) for each gradient run determined at the conclusion of step (b1), the difference between the maximum greyscale value and the minimum greyscale value of the pixels of the said gradient run is calculated;
- and the value of the said difference is associated with the said gradient run;
- (d1) a histogram is established giving, for each of the N.times.N pairs defined by (i) a greyscale value between 0 and N-1, and (ii) a value, between 0 and N-1, of the said differences calculated in step (c1), the number of pixels belonging to a gradient run and whose greyscale value and the value of the difference associated with the said gradient run are equal to the greyscale value in the said pair and to the value of the difference in the said pair respectively;
- (e1) from the histogram, is determined, among the N values of the said differences, a threshold dg0;
- (f1) the said machine-printed or handwritten text is retrieved in the form of pixels belonging to gradient runs, the value of the associated difference of which is above the said threshold dg0.
Dependencies are inferred from the claim text.