Go to DCV Architecture

The following diagram shows how sections connect to each other to form tasks:

In this article, we’ll discuss the section Document Detection which is the product-specific part of the 2nd section of a “Normalize-a-Document” task.

The 2nd section of a “Normalize-a-Document” task consists of Shared Detection and Document Detection.

Section 2.2 - Document Detection

The purpose of this section is to find the exact locations of “documents” on the image.

A document is an object that exhibit clear boundaries.

Constituting Stages

This section consists of multiple stages which forms a fixed and relatively complete set of workflow:

1. Contour-locating: to find contours that may be part of the document boundaries.
2. Line-locating: to find lines that may be part of the document boundaries.
3. Long-line-merging: to merge line segments into long lines.
4. Corner-locating: to find corners at intersections of long lines.
5. Candidate-quad-edge-locating: to find candidate edges that may be part of the quads.
6. Quad-locating: to find exact locations of documents.

Main Work Flow

The algorithm first finds the contour of the target item and extracts line segments from the contour. After a series of processing, the library calculates the corners, edges, and finally determines the position of the boundaries. Each boundary is output as a quadrilateral and the coordinates of its four vertices.

Output and Parameters

Each of these stages has its own output (known as an intermediate result) and usually a specific parameter that can regulate the operation:

The following parameter may affect the whole process:

• ContentType
• QuadrilateralDetectionModes