US 10,482,324 B2 · Method for analyzing a structured document likely to be deformed

Abstract

The invention relates to a method for analyzing a structured document likely to be deformed from a first image and a second image of the document, comprising steps of: matching first points of interest extracted in the first image with second points of interest of a reference image showing a model of the document, estimation of a first geometric transformation taking account of deformations of the structured document shown in the first image relative to the model from the matches, determination of at least one first region to be analyzed in the first image, by projection of at least one reference region of the reference image by means of the first transformation, analysis of the content of the first determined region, matching third points of interest extracted in the second image with fourth points of interest extracted in the first image, estimation, from the matches made in the preceding step, of a second geometric transformation taking account of deformations of the document shown in the second image relative to the document shown in the first image, estimation of a third geometric transformation taking account of deformations of the document shown in one of the two images, called target image, relative to the first model shown in the reference image, the third transformation dependent on the second transformation, determination of at least one second region to be analyzed in the target image by projection of the reference region of the reference image by means of the third geometric transformation, analysis of the content of the second determined region.

Claims15

Dependent claims

A method for analyzing content of a deformable structured document from of a first acquired (E1) image (A1) and a second acquired (E1′) image (A2) each showing the document, and comprising steps of:

matching (E6) first points of interest (P1A1) extracted in the first acquired image (A1) with second points of interest (P2M1) of a first reference image (M1) showing a first model of the document,
estimation (E8) of a first geometric transformation (TM1A1) taking account of deformations of the structured document shown in the first acquired image (A1) relative to the first model, from the matches made in the preceding step,
determination (E9) of at least one first region to be analyzed in the first acquired image (A1), by projection of at least one reference region of the first reference image by means of the first transformation (TM1A1),
analysis (E10) of the content of the first determined region,
wherein the method further comprises steps of:
matching (E12) third points of interest (P3A2) extracted in the second acquired image (A2) with fourth points of interest (P4A1) extracted in the first acquired image (A1),
estimation (E13), from the matches made in the preceding step (E12), of a second geometric transformation (TA1A2) taking account of deformations of the document shown in the second acquired image (A2) relative to the document shown in the first acquired image (A1),
estimation of a third geometric transformation (TM1A2) taking account of deformations of the document shown in the second acquired image (A2) relative to the first model shown in the first reference image (M1), wherein the third transformation is a composition of the first geometric transformation (TM1A1) and of the second geometric transformation (TA1A2),
determination (E14) of at least one second region to be analyzed in the second acquired image (A2) by projection of the reference region of the first reference image by means of the third geometric transformation,
analysis (E15) of the content of the second determined region.

7 more elements

A device for analyzing content of a deformable structured document, the device comprising a receiving interface for receiving a first image (A1) showing the document and a second image (A2) showing the document, and wherein the device further comprises an image-processing module (4) configured to:

match first points of interest (P1A1) extracted in the first acquired image (A1) with second points of interest (P2M2) of a first reference image (M1) showing a first model of the document,
estimate from a first geometric transformation (TM1A1) taking account of deformations of the structured document shown in the first acquired image relative to the first model, from the matches made in the preceding step,
determine at least one first region to be analyzed in the first acquired image (A1), by projection of at least one reference region of the first reference image by means of the first transformation (TM1A1),
analyze the content of the first determined region,
wherein the image-processing module (4) is further configured to:
match third points of interest (P3A2) extracted in the second acquired image (A2) with fourth points of interest (P4A1) extracted in the first acquired image (A1),
estimate, from the matches made in the preceding step, a second geometric transformation (TA1A2) taking account of deformations of the document shown in the second acquired image (A2) relative to the document shown in the first acquired image (A1),
estimate a third geometric transformation (TM1A2) taking account of deformations of the document shown in the second acquired image (A2) relative to the first model, wherein the third transformation is a composition of the first geometric transformation (TM1A1) and of the second geometric transformation (TA1A2),
determine at least one second region to be analyzed in the second acquired image (A2) by projection of the reference region of the first reference image by means of the third geometric transformation,
analyze the content of the second determined region.

7 more elements

match first points of interest (P1A1) extracted in the first acquired image (A1) with second points of interest (P2M2) of a first reference image (M1) showing a first model of the document,
estimate a first geometric transformation (TM1A1) taking account of deformations of the structured document shown in the first acquired image relative to the first model, from the matches made in the preceding step,
determine at least one first region to be analyzed in the first acquired image (A1), by projection of at least one reference region of the first reference image by means of the first transformation (TM1A1),
analyze the content of the first determined region,
wherein the image-processing module (4) is further configured to:
match third points of interest (P3A2) extracted in the second acquired image (A2) with fourth points of interest (P4A1) extracted in the first acquired image (A1),
estimate, from the matches made in the preceding step, a second geometric transformation (TA1A2) taking account of deformations of the document shown in the second acquired image (A2) relative to the document shown in the first acquired image (A1),
match the fifth points of interest (P5A2) extracted in the second acquired image (A2) with sixth points of interest (P6M2) of a second reference image (M2) showing a second model of the document,
estimate, from the matches made in the preceding step, a third geometric transformation (TM2A2) taking account of deformations of the document shown in the second acquired image (A2) relative to the second model,
estimate a fourth geometric transformation (TM1A1′) taking account of deformations of the document shown in the first acquired image (A1) relative to the first model, wherein the fourth transformation is a composition of the inverse (TA2A1) of the second transformation (TA1A2), of the third transformation (TM2A2), and of a fifth predetermined geometric transformation (TM1M2) taking account of deformations of the second model relative to the first model,
determine at least one second region to be analyzed in the first acquired image (A1) by projection of the reference region of the first reference image by means of the fourth geometric transformation,
analyze the content of the second determined region.

9 more elements

Dependencies are inferred from the claim text.

Method for analyzing a structured document likely to be deformed

Life of this patent

Method for analyzing a structured document likely to be deformed

Life of this patent

Abstract

Claims15