公开(公告)号：US11334967B2

公开(公告)日：2022-05-17

申请号：US16944097

申请日：2020-07-30

Applicant: Novatek Microelectronics Corp.

Inventor： Xin Huang ,  Fan Wang ,  Xi Tian

IPC: G06T5/00 ,  G06T5/50 ,  G06N3/08 ,  G06T3/40

Abstract: A method and an image processing device for generating an image filter are proposed. The method includes the following steps. A first image is received and inputted into a pre-trained convolutional neural network (CNN) model, and the image filter corresponding to the first image is obtained based on an output result of the pre-trained CNN model.

公开(公告)号：US20180184107A1

公开(公告)日：2018-06-28

申请号：US15674522

申请日：2017-08-11

Applicant: Novatek Microelectronics Corp.

Inventor： Pengyuan Hu ,  Xi Tian ,  Yen-Sung Chen ,  Yuan-Jia Du

IPC: H04N19/513

CPC classification number: H04N19/513 ,  H04N19/172 ,  H04N19/176 ,  H04N19/52 ,  H04N19/56

Abstract: The invention provides an apparatus and a method for motion estimation. The motion estimation apparatus includes a candidate set generation circuit and a motion vector (MV) determination circuit. The candidate set generation circuit selects to add motion vectors of adjacent blocks into a motion vector candidate set of a current block of a current frame. The current block and one of the adjacent blocks belong to the same row (or the same column), and the one of the adjacent blocks is a block of the current frame or a block of a previous frame. The motion vector determination circuit is coupled to the candidate set generation circuit to receive the motion vector candidate set. The motion vector determination circuit determines the motion vector of the current block of the current frame according to the motion vector candidate set.

公开(公告)号：US20220036510A1

公开(公告)日：2022-02-03

申请号：US16944097

申请日：2020-07-30

Applicant: Novatek Microelectronics Corp.

Inventor： Xin Huang ,  Fan Wang ,  Xi Tian

IPC: G06T5/00 ,  G06T5/50 ,  G06T3/40 ,  G06N3/08

Abstract: A method and an image processing device for generating an image filter are proposed. The method includes the following steps. A first image is received and inputted into a pre-trained convolutional neural network (CNN) model, and the image filter corresponding to the first image is obtained based on an output result of the pre-trained CNN model.

公开(公告)号：US10225587B1

公开(公告)日：2019-03-05

申请号：US15846220

申请日：2017-12-19

Applicant: NOVATEK Microelectronics Corp.

Inventor： Xi Tian ,  Xin Huang ,  Pengyuan Hu

IPC: H04N7/01 ,  H04N21/2343 ,  H04N21/24 ,  H04N21/234

Abstract: A motion estimation method for a frame rate converter includes generating a plurality of unidirectional motion vectors (MVs) based on a first original frame and a second original frame; classifying each of the plurality of unidirectional MVs as a reliable MV or a mismatch MV; mapping the reliable MV onto one of a plurality of blocks of an interpolation frame; classifying each of the plurality of blocks as a high confidence block or a low confidence block based on whether there is at least one reliable MV mapped onto the block; and determining an assigned MV for each of the high confidence block by selecting one of the at least one reliable MV mapped onto the high confidence block.

公开(公告)号：US10448043B2

公开(公告)日：2019-10-15

申请号：US15391864

申请日：2016-12-28

Applicant: NOVATEK Microelectronics Corp.

Inventor： Pengyuan Hu ,  Xi Tian ,  Yen-Sung Chen ,  Yuanjia Du

IPC: H04N19/52 ,  H04N19/513 ,  H04N19/105 ,  H04N19/14 ,  H04N19/56 ,  H04N19/176 ,  H04N19/172

Abstract: A motion estimation method and a motion estimator are provided. The motion estimation method includes the following steps of for each respective block of the current frame, selecting a candidate set associated with the current frame and a previous frame, and determining a motion vector according to the candidate set for the each respective block of the current frame. The candidate set includes at least one spatial candidate block in the current frame and temporal candidate blocks in the previous frame. The step of selecting the candidate set includes: selecting the spatial candidate block directly adjacent to the each respective block from the current frame and selecting the plurality of temporal candidate blocks directly adjacent to a reference block from the previous frame.