Method and apparatus for stitching images

1. An image stitching method, comprising:

obtaining a plurality of images at different viewpoints;

estimating homographies of the plurality of images;

generating an aligned image by aligning the plurality of images based on the homographies of each of the plurality of images;

obtaining a collective energy map by inputting the aligned image to a neural network;

estimating a stitching line of the aligned image based on the collective energy map; and

generating a stitched image by blending the aligned image based on the stitching line,

wherein the collective energy map combines an energy map using a gradient, an energy map using a motion, an energy map using object segmentation, and an energy map using distance transformation of object segmentation, and

wherein the image stitching method estimates the stitching line as a path having a minimum energy or a maximum energy.

2. The method of claim 1 , wherein the obtaining of the collective energy map comprises:

extracting feature information from the aligned image; and

calculating an energy for each pixel of the aligned image based on the feature information.

3. The method of claim 1 , wherein the neural network is configured to:

estimate an estimated energy map from a training image of a different viewpoint and learn to minimize a loss between the estimated energy map and a ground truth energy map.

4. The method of claim 3 , wherein the estimated energy map and the ground truth energy map comprises:

the energy map using the gradient of the training image of the different viewpoint, the energy map using the motion of an object between frames, or the energy map using the object segmentation.

5. An image stitching apparatus, comprising:

at least one processor; and

a memory,

wherein the processor is configured to:

obtain a plurality of images at different viewpoints;

estimate homographies of the plurality of images;

generate an aligned image by aligning the plurality of images based on the homographies of each of the plurality of images;

obtain a collective energy map by inputting the aligned image to a neural network;

estimate a stitching line of the aligned image based on the collective energy map; and

generate a stitched image by blending the aligned image based on the stitching line,

wherein the collective energy map combines an energy map using a gradient, an energy map using a motion, an energy map using object segmentation, and an energy map using distance transformation of object segmentation, and

wherein the image stitching apparatus estimates the stitching line as a path having a minimum energy or a maximum energy.