Method for reading document entries and addresses

• US 6,993,155 B1
• Filed: 02/01/2000
• Issued: 01/31/2006
• Est. Priority Date: 02/12/1999
• Status: Expired due to Term

• Current Assignee: Siemens Dematic AG (Siemens AG)
• Original Assignee: Siemens AG
• Inventors: Bayer, Thomas, Schäfer, Hartmut
• Primary Examiner(s): Ahmed, Samir
• Assistant Examiner(s): Le, Brian
• Application Number:
  US09/913,278
• Time in Patent Office: 2,191 Days
• Field of Search: 382/101, 382/102, 382/173, 382/175, 382176-177, 382/181, 382/190, 382224-229
• US Class Current: 382/101
• CPC Class Codes: G06F 18/21 Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
• CPC Class Codes: G06V 10/70 using pattern recognition or machine learning optical pattern recognition or electronic computations therefor G06V10/88
• CPC Class Codes: G06V 30/40 Document-oriented image-based pattern recognition

Abstract

The invention relates to a method for reading document entries and addresses. In this context, automatic reading is effected using the intermediate steps of layout analysis, classification of the segmented regions and interpretation of the results in a reader, and if necessary video coding. According to the invention, during the reading mode, when information has been clearly read automatically, the image data and the associated reading results and intermediate results, and/or, when information has been clearly ascertained using video coding, the image data and the reading results and intermediate results clearly ascertained in a second automatic reading process including the video coding results are stored for a current random sample. This current random sample is used to readapt the automatic reader.

Claims

1. A method for reading document entries and addresses using the following method steps

an image of the surface of documents or dispatches which contains the information to be read is recorded,

the information is automatically read in a reading mode in a reader using the following intermediate steps

layout analysis with

determination of the regions (ROI) which contain the information to be read,

segmentation into the areas which are to be classified,

classification of the segmented regions,

interpretation of the results from the preceding steps,

with intermediate results being stored, such that they can be picked off for subsequent intermediate steps, until a new image is read,

the image in question is video coded if a clear reading result has not been automatically ascertained within a prescribed time period,

the automatic reading steps being optimized in an adaptation phase using a random sample, comprising the images with the information to be read and the associated nominal reading results with their intermediate results,

wherein, during the reading mode, when information from selected images has been clearly read automatically, the reading results and intermediate results associated with the image data of the respective image are stored for a current random sample, and/or, when information from the selected images has been clearly ascertained using video coding, the image data are read once again in a second automatic reading process, having the same functionality, and the associated information ascertained by video coding is included, and in the case of a clear reading result, the reading results and intermediate results are likewise stored, assigned to the associated image data, for the current random sample, and in that the current random sample is used to readapt the automatic reader and the images to be included in the random sample are selected such that the random sample contains a stipulated ratio of automatically read images and video coded images.

2. The method according to claim 1, wherein the reading process for the second automatic reading is carried out with no time limitations.

3. The method according to claim 1, further comprising the steps of determining unsuitable reading and intermediate results and excluding such results from the random sample.

4. The method according to claim 1, wherein, for each image read successfully using video coding, the intermediate results for the first automatic reading are compared with the associated intermediate results for the second automatic reading, and significant discrepancies are summed as evaluation errors in the respective intermediate steps during the first automatic reading to form a statistic.

5. The method according to claim 4, wherein only the intermediate steps whose ascertained frequency of error is above a stipulated value during a particular tine period are readapted.

6. The method according to claim 1, wherein the reading rate for the first automatic reading is monitored, and if the reading rate is below a particular threshold value over a stipulated time period, the reader or the intermediate steps are readapted using the current random sample.

7. The method according to claim 1 wherein the reading rate is ascertained on an automatic reader having the same functionality using a random sample in the case of the previous and the readapted parameters, and if the reading rate is improved by the readapted parameters, the automatic reader is reinitialized.

8. The method according to claim 1, wherein the images to be included in the random sample are selected such that the random sample contains a particular number of newly read images having clearly read information within a stipulated time period.

9. The method according to claim 1, wherein the random sample data for the selected, currently read images are inserted into the original random sample, and if a maximum number of random sample elements is exceeded, the oldest elements in each case are removed.

10. The method according to claim 1, wherein the original random sample is replaced by a current random sample.

11. A method for reading document entries and addresses using the following method steps

an image of the surface of documents or dispatches which contains the information to be read is recorded,

the information is automatically read in a reader using the following intermediate steps

layout analysis with

determination of the regions (ROI) which contain the information to be read,

segmentation into the areas which are to be classified,

classification of the segmented regions,

interpretation of the results from the preceding steps,

with intermediate results being stored, such that they can be picked off for subsequent intermediate steps, until a new image is read,

the image in question is video coded if a clear reading result has not been automatically ascertained within a prescribed time period,

the automatic reading steps being optimized in an upstream adaptation phase using a random sample, comprising the images with the information to be read and the associated desired reading results with their intermediate result,

characterized in that, during a reading mode, when information from selected images has been clearly read automatically, the reading results and intermediate results associated with the image data of the respective image are stored for a current random sample, and/or, when information from the selected images has been clearly ascertained using video coding, the image data are read once again in a second automatic reading process, having the same functionality, and the associated information ascertained by video coding is included, and in the case of a clear reading result, the reading results and intermediate results are likewise stored, assigned to the associated image data, for the current random sample, and in that the current random sample is used to readapt the automatic reader, and wherein the reading rate is ascertained on an automatic reader having the same functionality using a random sample in the case of the previous and the readapted parameters, and if the reading rate is improved by the readapted parameters, the automatic reader is reinitialized.

12. The method according to claim 11, wherein the reading process for the second automatic reading is carried out with no time limitations.

13. The method according to claim 11, further comprising the steps of determining unsuitable reading and intermediate results and excluding such results from the random sample.

14. The method according to claim 11, wherein, for each image read successfully using video coding, the intermediate results for the first automatic reading are compared with the associated intermediate results for the second automatic reading, and significant discrepancies are summed as evaluation errors in the respective intermediate steps during die first automatic reading to form a statistic.

15. The method according to claim 14, wherein only the intermediate steps whose ascertained frequency of error is above a stipulated value during a particular time period are readapted.

16. The method according to claim 11, wherein the reading rate for the first automatic reading is monitored, and if the reading rate is below a particular threshold value over a stipulated time period, the reader or the intermediate steps are readapted using the current random sample.