Learning Filters For Enhancing The Quality Of Block Coded Still And Video Images
First Claim
1. A method for block-wise encoding still or video images comprising the steps of:
- a) encoding input image data composed of pixels having original pixel values using a block-wise encoder to obtain encoded image data;
b) decoding at least a subset of the encoded image data to obtain a decoded image composed of pixels having decoded image values;
c) identifying in the decoded image a plurality of coding block boundaries;
d) forming training data comprising a plurality of cross-boundary filter support blocks (CBFSBs), each CBFSB comprising a group of N>
1 adjacent pixels of the decoded image including adjacent pixels on both sides of a block boundary;
e) generating a set of classification rules based on the training data for classifying each of the CBFSBs into at least one of a plurality of classes, so that each class is adaptively defined by a respective set of the classification rules;
f) for each class defined in step (e), generating a set of filter coefficients associated with the class and defining a class-optimized filter that provides a prediction value of a selected CBFSB pixel in dependence upon decoded values of CBFSB pixels based on an original value of the selected CBFSB pixel for each CBFSB in said class; and
,g) providing the sets of filter coefficients for each class and the classification rules as an output for subsequent adaptive class-based filtering of the decoded image data for suppressing blocking artefacts.
1 Assignment
0 Petitions
Accused Products
Abstract
The invention relates to a method and an apparatus for reducing blocking artifacts in block-wise coding of still and video images. A learning filter generator is provided at the image encoder for generating a set of filters and associated filtering rules for filtering cross-boundary image patterns based on representative original and decoded training images using a supervised machine learning algorithm. An adaptive filter at the image decoder receives the generated filters and associated filtering rules and performs locally adaptive filtering in accordance with the received filtering rules.
-
Citations
20 Claims
-
1. A method for block-wise encoding still or video images comprising the steps of:
-
a) encoding input image data composed of pixels having original pixel values using a block-wise encoder to obtain encoded image data; b) decoding at least a subset of the encoded image data to obtain a decoded image composed of pixels having decoded image values; c) identifying in the decoded image a plurality of coding block boundaries; d) forming training data comprising a plurality of cross-boundary filter support blocks (CBFSBs), each CBFSB comprising a group of N>
1 adjacent pixels of the decoded image including adjacent pixels on both sides of a block boundary;e) generating a set of classification rules based on the training data for classifying each of the CBFSBs into at least one of a plurality of classes, so that each class is adaptively defined by a respective set of the classification rules; f) for each class defined in step (e), generating a set of filter coefficients associated with the class and defining a class-optimized filter that provides a prediction value of a selected CBFSB pixel in dependence upon decoded values of CBFSB pixels based on an original value of the selected CBFSB pixel for each CBFSB in said class; and
,g) providing the sets of filter coefficients for each class and the classification rules as an output for subsequent adaptive class-based filtering of the decoded image data for suppressing blocking artefacts. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 18, 19, 20)
-
-
17. An image encoder for encoding still or video images, comprising:
-
a block-wise encoder for generating encoded image data from input image data comprised of pixels having original pixel values using coding blocks having coding block boundaries; a block-wise decoder operatively connected to the block-wise encoder for receiving at least a subset of the encoded image data and for generating therefrom a decoded image comprised of pixels having decoded pixel values; a training data generator for forming a training dataset comprising a plurality of cross-boundary filter support blocks (CBFSB), each CBFSB comprising a group of N>
1 adjacent pixels of the decoded image including adjacent pixels on both sides of a block boundary;a learning classifier operatively connected to the input port and the training data generator for receiving the input image and the corresponding decoded image for generating therefrom a set of classification rules for classifying each of the CBFSBs into at least one of a plurality of classes that are thereby adaptively defined, and at least one set of class-optimized filter coefficients for each adaptively defined class so that filtered values of CBFSB pixels obtained using said filter coefficients approximate original pixel values of respective pixels for all CBFSB associated with said class; and
,a data port coupled to the block encoder and the learning classifier for providing the encoded image data, the classification rules and the set at least one set of the class-optimized filter coefficients as an output for subsequent decoding and class-based pixel filtering for reducing blocking artefacts.
-
Specification