Image face changing method based on local occlusion condition

Abstract

The invention provides an image face changing method based on a local occlusion condition. An input face image and a target face image are described by two-dimensional feature points, related face areas are divided on the input image and the target image. A pixel feature function corresponding to the face areas is established, and seamless fusion is performed in the face area migration process toeliminate splicing boundaries. A final output image is synethisized through Laplacian pyramid reconstruction by using an idea of image layered superposition. The finally synthesized image has no obvious modification splicing trace. Compared with a Poisson fusion method commonly used for image face changing at present, image effect of an edge of the shielding object is well restored, especially when the shielding object has projection shadows in the face area, the details of the edge area of the shielding object are obviously different from those of a Poisson fusion method, and the image effectof the whole image after face changing is harmonious and natural and has no obvious defects.

Description

technical field

[0001] The invention relates to the technical fields of image processing and computer graphics, and mainly relates to a face-changing method of a target image based on partial occlusion. Background technique

[0002] Image face-swapping technology refers to automatically transplanting the user's face in the input image to the target image, replacing the specified face in the target image, so that the image's face is automatically fused into the target image. The process of image face-changing requires keeping the facial appearance and expression features of the input image unchanged, but it must match the pose of the target face, and the final synthesized face image must have a natural overall effect, that is, it is required that the facial features after face-changing The skin color, makeup, and lighting environment of the input face can be matched with the target image. This technology can make the user who inputs the image have a sense of substitution and...

Examples

Embodiment Construction

[0074] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0075] The resolution of the test image is selected as 800×600, and the corresponding hardware test environment is memory 16G, CPU 4.0Hz, and the development platform Matlab.

[0076] Step 1: Preprocessing the input image and the target image;

[0077] For the input image I S and target image I T Mark the occluders in , and then use the marked points to extract the occluders, that is, to divide the occluders according to their contours, and respectively divide the input image I S and target image I T Carry out occlusion and face area segmentation. When there is no occlusion, no occlusion segmentation is performed and only face area segmentation is performed. When the face image has image occlusions other than the face itself, the shape of the occlusion is used as the segmentation contour , the occluder is segmented, and thus the occluder area and th...

technical field

[0001] The invention relates to the technical fields of image processing and computer graphics, and mainly relates to a face-changing method of a target image based on partial occlusion. Background technique

[0002] Image face-swapping technology refers to automatically transplanting the user's face in the input image to the target image, replacing the specified face in the target image, so that the image's face is automatically fused into the target image. The process of image face-changing requires keeping the facial appearance and expression features of the input image unchanged, but it must match the pose of the target face, and the final synthesized face image must have a natural overall effect, that is, it is required that the facial features after face-changing The skin color, makeup, and lighting environment of the input face can be matched with the target image. This technology can make the user who inputs the image have a sense of substitution and...

Examples

Embodiment Construction

[0074] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0075] The resolution of the test image is selected as 800×600, and the corresponding hardware test environment is memory 16G, CPU 4.0Hz, and the development platform Matlab.

[0076] Step 1: Preprocessing the input image and the target image;

[0077] For the input image I S and target image I T Mark the occluders in , and then use the marked points to extract the occluders, that is, to divide the occluders according to their contours, and respectively divide the input image I S and target image I T Carry out occlusion and face area segmentation. When there is no occlusion, no occlusion segmentation is performed and only face area segmentation is performed. When the face image has image occlusions other than the face itself, the shape of the occlusion is used as the segmentation contour , the occluder is segmented, and thus the occluder area and th...

Images