Systems and methods for computationally distinguishing handwritten pencil marks from preprinted marks in a scanned document
View Patent ↗A method, system, and algorithm detects graphite responses in optical mark recognition documents using inexpensive visible-light scanners. The invention is invariant to the numerous background colors and artwork in typical optical mark recognition documents. A superposition of a graphite response image and a background image is modeled. The background image in turn is modeled as a superposition of screening, artwork, lines, and machine text components. A sequence of image processing operations and a pattern recognition algorithm is used to isolate the graphite response image from a test document by systematically removing the components of the background image. The image processing operations consist of grayscale and color segmentation and the application of the Hough transform. Components that are not removed by image processing operations are identified by calculating their texture features, and removed using a multivariate Gaussian classifier.
1. A method for computationally distinguishing hand-written marks from pre-printed artwork in a response sheet marked with pre-printed elements and hand-written elements, comprising:
scanning a sheet containing only pre-printed elements to generate a first classification pixel file;
identifying one or more connected pixel groups from data corresponding to the first classification pixel file;
computationally detecting one or more features of the connected pixel groups of the first classification pixel file, and storing the detected features of the connected pixel groups of the first classification pixel file as a first classification set;
scanning a sheet containing only hand-written elements to generate a second classification pixel file;
identifying one or more connected pixel groups from data corresponding to the second classification pixel file;
computationally detecting one or more features of the connected pixel groups of the second classification pixel file, and storing the detected features of the connected pixel groups of the second classification pixel file, as a second classification set;
scanning the response sheet marked with pre-printed components and hand-written components to generate a response pixel file;
identifying one or more connected pixel groups from data corresponding to the response pixel file;
for each connected pixel group of the response pixel file, computationally detecting from data corresponding to the response pixel file one or more features of the connected pixel group; and
computationally determining if the connected pixel group of the response pixel file corresponds to a pre-printed element or a hand-written element by comparing the features detected from the response pixel file with the features stored in the first classification set and the features stored in the second classification set.
2. The method of claim 1 , wherein the detected features of the first classification set, the detected features of the second classification set, and the detected features of the response pixel file comprise one or more textural features selected from the group comprising: entropy, skewness, uniformity, mean, standard deviation, smoothness, kurtosis, root mean square, variance, median absolute deviation, range, and interquartile range.
3. The method of claim 1 , further comprising the step of computationally removing data values corresponding to vertical and horizontal lines from the response pixel file.
4. The method of claim 3 , wherein the step of computationally removing data values corresponding to vertical and horizontal lines from the response pixel file comprises applying a Hough transform to the values of the response pixel file.
5. The method of claim 1 , further comprising the step of converting the response pixel file to a grayscale image.
6. The method of claim 1 , further comprising, after determining if the connected pixel groups of the response pixel file corresponds to a pre-printed component or a hand-written component, suppressing the connected pixel groups corresponding to pre-printed components and storing, printing, or displaying only connected pixel groups corresponding to the hand-written components.
7. The method of claim 1 , wherein the method is invariant of the color or colors of the pre-printed elements.
8. A system for computationally distinguishing hand-written marks from pre-printed marks in a response sheet marked with pre-printed elements and hand-written elements, comprising:
storage media configured for storing data corresponding to one or more pixel files, said storage media storing:
a first classification set comprising computationally-detected features of connected pixel groups of data corresponding to a first classification pixel file generated by scanning a sheet containing only pre-printed components; and
a second classification set comprising computationally-detected features of connected pixel groups of data corresponding to a second classification pixel file generated by scanning a sheet containing only hand-written components; and
an element determination module comprising a computer configured to computationally detect features from connected pixel groups of data corresponding to a response pixel file generated by scanning a response sheet marked with pre-printed elements and hand-written elements and to compare computationally detected features of a scanned response sheet with features of the first classification set and features of the second classification set to determine if connected pixel groups of the scanned response sheet correspond to pre-printed elements or hand-written elements.
9. The system of claim 8 , further comprising a scanner for scanning sheets containing pre-printed marks and/or hand-written marks and generating a pixel file for the scanned sheet.
10. The system of claim 8 , wherein the features detected from the connected pixel groups of the response pixel file comprise one or more textural features selected from the group comprising: entropy, skewness, uniformity, mean, standard deviation, smoothness, kurtosis, root mean square, variance, median absolute deviation, range, and interquartile range.
11. The system of claim 8 , further comprising a computer configured to computationally remove data values corresponding to vertical and horizontal lines from the response pixel file.
12. The system of claim 11 , wherein the computer is configured to computationally remove data values corresponding to vertical and horizontal lines from the response pixel file by applying a Hough transform to the values of the response pixel file.
13. The system of claim 8 , further comprising a computer configured to convert the response pixel file to a grayscale image.
14. The system of claim 8 , further comprising a computer configured to suppress connected pixel groups corresponding to the pre-printed elements of the response sheet and to store, print, or display only connected pixel groups corresponding to the hand-written elements of the response sheet.
15. The system of claim 8 , wherein the element determination module is configured to determine if connected pixel elements of the scanned response sheet correspond to pre-printed elements or hand-written elements without regard to the color or colors of the pre-printed elements.