公开(公告)号：US10250907B2

公开(公告)日：2019-04-02

申请号：US15326882

申请日：2015-01-04

Applicant: Tsinghua University

Inventor： Pin Tao ,  Lixin Feng

IPC: H04N19/593 ,  H04N19/61 ,  H04N19/88 ,  H04N19/56

Abstract: Provided are an intra-frame pixel prediction method, encoding method and decoding method, and an intra-frame pixel prediction device. The intra-frame pixel prediction method comprises: defining a target template of a pixel to be predicted currently; comparing the target template with candidate templates in a search region of the frame, and determining, from the candidate templates, at least one matching template matching the target template; and based on the at least one matching template, determining a prediction value of the pixel to be predicted currently. In the intra-frame pixel prediction method in the embodiments of the present invention, the template matching is conducted by taking a pixel point as a unit, so that a pixel can be predicted accurately.

公开(公告)号：US11057621B2

公开(公告)日：2021-07-06

申请号：US16531513

申请日：2019-08-05

Applicant: Tsinghua University

Inventor： Pin Tao ,  Lixin Feng ,  Zheng Wang

IPC: H04N19/11 ,  H04N19/61 ,  H04N19/86 ,  H04N19/14 ,  H04N19/124 ,  H04N19/13 ,  H04N19/182 ,  H04N19/91

Abstract: An image encoding method, decoding method and corresponding device, and intraframe pixel prediction method. The image encoding method comprises: a step of pixel segmentation, segmenting pixels in an image frame; a step of pixel value prediction, respectively performing a pixel value prediction process to obtain a first prediction value for each of pixels in a current block to be encoded; a step of residual calculation, calculating a residual for each of the pixels in the current block to be encoded; a step of discrete cosine transform, quantization and entropy encoding, performing discrete cosine transform, quantization and entropy encoding with respect to a residual block corresponding to the current block to be encoded obtained in the residual calculation step, wherein the quantized residual block is used in a pixel value reconstruction step at an encoding end, and the entropy encoded residual block is sent to a decoding end; and a pixel value reconstruction step, reconstructing a pixel value of the previously-encoded pixel, wherein the re-constructed pixel value is used in the pixel value prediction step. The residual calculation step comprises: calculating a first residual with respect to a pixel on the basis of the first prediction value of the pixel predicted in the pixel value prediction step and an actual value of the pixel; determining whether the first residual of the pixel is a normal residual or an abnormal residual; correcting the abnormal residual if the first residual of the pixel is an abnormal residual, the correction including recording a position of which the abnormal residual of the pixel has occurred; re-predicting the pixel to obtain a second prediction value and a corresponding second residual, and recording the re-prediction information, such that the pixel value reconstruction step of the encoding end and the decoding end can both reconstruct and decode the pixel value of the pixel on the basis of the position of which the abnormal residual of the pixel has occurred and the re-prediction information, wherein the normal residual block of the current block is formed by the residual of the pixel of the current block to be encoded after undergoing the abnormal residual correction, and is provided for further processing in the discrete cosine transform, quantization and entropy encoding step; separately encoding the position of which the abnormal residual of the pixel has occurred and the re-prediction information, wherein the encoding process is lossless, and the encoded pixel position and the re-prediction information are used in the pixel value reconstruction step and sent to the decoding end. Pixels can be re-arranged before segmentation, thereby avoiding error diffusion. In the pixel value prediction process, a prediction algorithm can be predicted first, and the pixel value is then predicted according to the determined prediction algorithm. By classifying the pixel residual into the normal residual and the abnormal residual, as well as by performing a re-prediction of the pixel of the abnormal residual, the accuracy of image encoding/decoding can be improved.

公开(公告)号：US20190356914A1

公开(公告)日：2019-11-21

申请号：US16531513

申请日：2019-08-05

Applicant: Tsinghua University

Inventor： Pin Tao ,  Lixin Feng ,  Zheng Wang

IPC: H04N19/11 ,  H04N19/14 ,  H04N19/91 ,  H04N19/182 ,  H04N19/61 ,  H04N19/124 ,  H04N19/86 ,  H04N19/13

Abstract: An image encoding method, decoding method and corresponding device, and intraframe pixel prediction method. The image encoding method comprises: a step of pixel segmentation, segmenting pixels in an image frame; a step of pixel value prediction, respectively performing a pixel value prediction process to obtain a first prediction value for each of pixels in a current block to be encoded; a step of residual calculation, calculating a residual for each of the pixels in the current block to be encoded; a step of discrete cosine transform, quantization and entropy encoding, performing discrete cosine transform, quantization and entropy encoding with respect to a residual block corresponding to the current block to be encoded obtained in the residual calculation step, wherein the quantized residual block is used in a pixel value reconstruction step at an encoding end, and the entropy encoded residual block is sent to a decoding end; and a pixel value reconstruction step, reconstructing a pixel value of the previously-encoded pixel, wherein the re-constructed pixel value is used in the pixel value prediction step. The residual calculation step comprises: calculating a first residual with respect to a pixel on the basis of the first prediction value of the pixel predicted in the pixel value prediction step and an actual value of the pixel; determining whether the first residual of the pixel is a normal residual or an abnormal residual; correcting the abnormal residual if the first residual of the pixel is an abnormal residual, the correction including recording a position of which the abnormal residual of the pixel has occurred; re-predicting the pixel to obtain a second prediction value and a corresponding second residual, and recording the re-prediction information, such that the pixel value reconstruction step of the encoding end and the decoding end can both reconstruct and decode the pixel value of the pixel on the basis of the position of which the abnormal residual of the pixel has occurred and the re-prediction information, wherein the normal residual block of the current block is formed by the residual of the pixel of the current block to be encoded after undergoing the abnormal residual correction, and is provided for further processing in the discrete cosine transform, quantization and entropy encoding step; separately encoding the position of which the abnormal residual of the pixel has occurred and the re-prediction information, wherein the encoding process is lossless, and the encoded pixel position and the re-prediction information are used in the pixel value reconstruction step and sent to the decoding end. Pixels can be re-arranged before segmentation, thereby avoiding error diffusion. In the pixel value prediction process, a prediction algorithm can be predicted first, and the pixel value is then predicted according to the determined prediction algorithm. By classifying the pixel residual into the normal residual and the abnormal residual, as well as by performing a re-prediction of the pixel of the abnormal residual, the accuracy of image encoding/decoding can be improved.

公开(公告)号：US10375391B2

公开(公告)日：2019-08-06

申请号：US15579000

申请日：2015-06-04

Applicant: Tsinghua University

Inventor： Pin Tao ,  Lixin Feng ,  Zheng Wang

IPC: H04N19/11 ,  H04N19/13 ,  H04N19/14 ,  H04N19/61 ,  H04N19/86 ,  H04N19/91 ,  H04N19/124 ,  H04N19/182

Abstract: An image encoding method, decoding method and corresponding device, and intraframe pixel prediction method are disclosed. The image encoding method comprises: pixel segmentation, segmenting pixels in an image frame; pixel value prediction, respectively performing a pixel value prediction process to obtain a first prediction value for each of pixels in a current block to be encoded; a step of residual calculation, calculating a residual for each of the pixels in the current block to be encoded; a step of discrete cosine transform, quantization and entropy encoding, performing discrete cosine transform, quantization and entropy encoding with respect to a residual block corresponding to the current block to be encoded obtained in the residual calculation step, and the entropy encoded residual block is sent to a decoding end; and a pixel value reconstruction step, reconstructing a pixel value of the previously-encoded pixel.

公开(公告)号：US20180146192A1

公开(公告)日：2018-05-24

申请号：US15579000

申请日：2015-06-04

Applicant: Tsinghua University

Inventor： Pin Tao ,  Lixin Feng ,  Zheng Wang

IPC: H04N19/11 ,  H04N19/124 ,  H04N19/13 ,  H04N19/91 ,  H04N19/182

CPC classification number: H04N19/11 ,  H04N19/124 ,  H04N19/13 ,  H04N19/14 ,  H04N19/182 ,  H04N19/61 ,  H04N19/86 ,  H04N19/91

Abstract: An image encoding method, decoding method and corresponding device, and intraframe pixel prediction method. The image encoding method comprises: a step of pixel segmentation, segmenting pixels in an image frame; a step of pixel value prediction, respectively performing a pixel value prediction process to obtain a first prediction value for each of pixels in a current block to be encoded; a step of residual calculation, calculating a residual for each of the pixels in the current block to be encoded; a step of discrete cosine transform, quantization and entropy encoding, performing discrete cosine transform, quantization and entropy encoding with respect to a residual block corresponding to the current block to be encoded obtained in the residual calculation step, wherein the quantized residual block is used in a pixel value reconstruction step at an encoding end, and the entropy encoded residual block is sent to a decoding end; and a pixel value reconstruction step, reconstructing a pixel value of the previously-encoded pixel, wherein the reconstructed pixel value is used in the pixel value prediction step. The residual calculation step comprises: calculating a first residual with respect to a pixel on the basis of the first prediction value of the pixel predicted in the pixel value prediction step and an actual value of the pixel; determining whether the first residual of the pixel is a normal residual or an abnormal residual; correcting the abnormal residual if the first residual of the pixel is an abnormal residual, the correction including recording a position of which the abnormal residual of the pixel has occurred; re-predicting the pixel to obtain a second prediction value and a corresponding second residual, and recording the re-prediction information, such that the pixel value reconstruction step of the encoding end and the decoding end can both reconstruct and decode the pixel value of the pixel on the basis of the position of which the abnormal residual of the pixel has occurred and the re-prediction information, wherein the normal residual block of the current block is formed by the residual of the pixel of the current block to be encoded after undergoing the abnormal residual correction, and is provided for further processing in the discrete cosine transform, quantization and entropy encoding step; separately encoding the position of which the abnormal residual of the pixel has occurred and the re-prediction information, wherein the encoding process is lossless, and the encoded pixel position and the re-prediction information are used in the pixel value reconstruction step and sent to the decoding end. Pixels can be re-arranged before segmentation, thereby avoiding error diffusion. In the pixel value prediction process, a prediction algorithm can be predicted first, and the pixel value is then predicted according to the determined prediction algorithm. By classifying the pixel residual into the normal residual and the abnormal residual, as well as by performing a re-prediction of the pixel of the abnormal residual, the accuracy of image encoding/decoding can be improved.

公开(公告)号：US20170223378A1

公开(公告)日：2017-08-03

申请号：US15326882

申请日：2015-01-04

Applicant: Tsinghua University

Inventor： Pin Tao ,  Lixin Feng

IPC: H04N19/593 ,  H04N19/88

CPC classification number: H04N19/593 ,  H04N19/56 ,  H04N19/61 ,  H04N19/88

Abstract: Provided are an intra-frame pixel prediction method, encoding method and decoding method, and an intra-frame pixel prediction device. The intra-frame pixel prediction method comprises: defining a target template of a pixel to be predicted currently; comparing the target template with candidate templates in a search region of the frame, and determining, from the candidate templates, at least one matching template matching the target template; and based on the at least one matching template, determining a prediction value of the pixel to be predicted currently. In the intra-frame pixel prediction method in the embodiments of the present invention, the template matching is conducted by taking a pixel point as a unit, so that a pixel can be predicted accurately. According to the pixel encoding method and decoding method in the embodiments of the present invention comprising the pixel rearranging technology and the intra-frame prediction technology, the method for conducting intra-frame prediction by taking a pixel as a unit can be used in combination with the traditional technologies of taking a block as a unit, such as transformation, quantification and encoding, and at the same time, the problem of the spread of prediction errors can be overcome or reduced.