Inter prediction mode-based image processing method and apparatus therefor
Disclosed is a method for processing an image based on an inter prediction and an apparatus for the same. Particularly, the method may include generating a bi-directional predictor of each pixel of a current block by performing a bi-directional inter prediction based on a motion vector of the current block; deriving a first motion vector from a window area of a specific size in a first reference picture for a current pixel, and deriving a second motion vector from a window area of a specific size in a second reference picture for the current pixel; deriving a first predictor for the current pixel using the first motion vector, and deriving a second predictor for the current pixel using the second motion vector; and generating a third predictor of each pixel of the current block by performing a weighted sum of the first predictor and the second predictor.
1. A method for processing an image based on an inter prediction by an apparatus, comprising:
generating a bi-directional predictor of each pixel of a current block by performing a bi-directional inter prediction based on a motion vector of the current block;
deriving a first motion vector from a window area of a specific size in a first reference picture for a current pixel, and deriving a second motion vector from a window area of a specific size in a second reference picture for the current pixel,
wherein a horizontal directional component of the first motion vector is configured as 0, when a sum of squared values of the horizontal directional component of a gradient for each pixel is smaller than a specific threshold value,
wherein a vertical directional component of the first motion vector is configured as 0, when a sum of squared values of the vertical directional component of a gradient for each pixel is smaller than a specific threshold value,
wherein a horizontal directional component of the second motion vector is configured as 0, when a sum of squared values of the horizontal directional component of a gradient for each pixel is smaller than a specific threshold value, and
wherein a vertical directional component of the second motion vector is configured as 0, when a sum of squared values of the vertical directional component of a gradient for each pixel is smaller than a specific threshold value;
deriving a first predictor for the current pixel using the first motion vector, and deriving a second predictor for the current pixel using the second motion vector; and
generating a third predictor of each pixel of the current block by performing a weighted sum of the first predictor and the second predictor.
2. The method of claim 1 , wherein a vertical directional component and a horizontal directional component of the gradient are obtained by using a difference value between a horizontal direction pixel value and a vertical direction pixel value with a pixel of which coordinate is collocated with the current pixel in a center in the first reference picture or the second reference picture.
3. The method of claim 1 , wherein a vertical directional component and a horizontal directional component of the gradient are obtained by applying interpolation filters of a horizontal direction and a vertical direction with a pixel of which coordinate is collocated with the current pixel in a center in the first reference picture or the second reference picture.
4. The method of claim 1 , wherein the first predictor is generated by using a gradient for each pixel and the first motion vector in the window area in the first reference picture, and
wherein the second predictor is generated by using a gradient for each pixel and the second motion vector in the window area in the second reference picture.
5. An apparatus for processing an image based on an inter prediction, comprising:
a memory; and
a processor configured to generate a bi-directional predictor of each pixel of a current block by performing a bi-directional inter prediction based on a motion vector of the current block,
derive a first motion vector from a window area of a specific size in a first reference picture for a current pixel, and derive a second motion vector from a window area of a specific size in a second reference picture for the current pixel,
wherein a horizontal directional component of the first motion vector is configured as 0, when a sum of squared values of the horizontal directional component of a gradient for each pixel is smaller than a specific threshold value,
wherein a vertical directional component of the first motion vector is configured as 0, when a sum of squared values of the vertical directional component of a gradient for each pixel is smaller than a specific threshold value,
wherein a horizontal directional component of the second motion vector is configured as 0, when a sum of squared values of the horizontal directional component of a gradient for each pixel is smaller than a specific threshold value, and
wherein a vertical directional component of the second motion vector is configured as 0, when a sum of squared values of the vertical directional component of a gradient for each pixel is smaller than a specific threshold value,
derive a first predictor for the current pixel using the first motion vector, and derive a second predictor for the current pixel using the second motion vector, and
generate a third predictor of each pixel of the current block by performing a weighted sum of the first predictor and the second predictor.