Image seamless copying method

Abstract

The invention discloses a method for copying a selected region in a source image into a target image in a seamless manner. The method comprises the following steps: selecting the copied region in the source image, and carrying out pre-treatment to approximate the copied region as a polygonal region; carrying out self-adaptive triangulation on the polygonal region, and calculating tuning coordinate of each vertex of a triangular mesh about a polygon boundary; carrying out hierarchical sampling and re-normalization on the tuning coordinates according to the relative position attribute of each vertex of the triangular mesh about the polygon boundary; calculating the color difference of each vertex of a polygon between corresponding pixels which are respectively positioned in the target image and the source image according to the position of the copied polygonal region in the target image; calculating the color difference of the vertices of the triangular mesh according to the tuning coordinate and the corresponding color difference of each vertex of the polygon; and using programmable graphics hardware to sequentially draw the target image and the triangular mesh with vertex attributes, and obtaining a final fused image.

Description

technical field [0001] The invention relates to a method for seamless image duplication, and the duplication result can be fed back to the user in real time. Background technique [0002] In image editing, a common operation is to use part or all of the source image as the copied area, and copy it seamlessly to a specific area of ​​the target image, so that the final fused image does not exist on the boundary of the copied area. Visually perceivable and obvious boundaries, so it is called seamless copy. This technology is widely used in image repair, image background replacement and other fields. [0003] The traditional seamless replication technology establishes an optimization equation in the replicated area, and usually uses the Poisson equation with boundary constraints to describe this requirement. However, solving the Poisson equation is a very time-consuming operation, and the result cannot be interactively fed back to the user (for related techniques, please refer...

AI Technical Summary

Problems solved by technology

However, solving the Poisson equation is a very time-consuming operation, and the results cannot be interactively fed back to the user (for related techniques, please refer to Pérez P., Gangnet M., Blake A., 2003.Poisson image editing)

With the development of GPU technology, researchers have proposed some Poisson equation solvers based on parallel architecture, but failed to fundamentally solve the efficiency problem

The method based on Mean-value coordinate interpolation introduces coordinate pre-computation with a small cost, and only performs the vector dot product operation during image copying, thereby saving time-consuming equation solving, so it can feedback the copying results to the user in real time; However, when the copied area is concave, this technique cannot maintain the correct gradient field, and the obtained coordinate vector cannot correctly indicate that the internal pixels are affected by the color difference of each vertex at the border of the copied area, resulting in the fusion result of the image The color cross effect between the two lobes of the concave shape, such that there is a visible transition in the resulting image

Images