Scene detection method, chip, electronic device, and storage medium

Patent metadata

Field	Value
Publication number	US-2021407104-A1
Application number	US-202117304959-A
Country	US
Kind code	A1
Filing date	Jun 29, 2021
Priority date	Jun 30, 2020
Publication date	Dec 30, 2021
Grant date	—

Abstract

The present disclosure relates to a scene detection method, a chip, an electronic device, and a storage medium, the present disclosure detects periodic scenes, which helps to improve the accuracy of motion estimation.

First claim

What is claimed is:

1 . A scene detection method, comprising:
calculating a match error between a current central block and a surrounding block within a search range centered on the current central block; obtaining a first match error reference maximum value based on the central block along a row direction as a horizontal prediction depth; obtaining a second match error reference maximum value based on the central block along a column direction as a vertical prediction depth; setting a first horizontal threshold and a second horizontal threshold based on the horizontal prediction depth, and the first horizontal threshold is greater than the second horizontal threshold; along the row direction of the search range, based on a relative magnitude of the match error of the block with the first horizontal threshold and with the second horizontal threshold, judging the number of periodic changes of match error as a first numerical value; setting a first vertical threshold and a second vertical threshold based on the vertical prediction depth, and the first vertical threshold is greater than the second vertical threshold; along the column direction of the search range, based on a relative magnitude of the match error of the block with the first vertical threshold and with the second vertical threshold, judging the number of periodic changes of match error as a second numerical value; and determining whether the current central block is located in a periodic scene or not based on the first numerical value and the second numerical value.

2 . The scene detection method of claim 1 , wherein obtaining the first match error reference maximum value comprises:
obtaining a first number of blocks with successively increasing match errors based on the central block along the row direction as a horizontal prediction radius; and obtaining the match error of the block corresponding to the horizontal prediction radius at a distance from the central block along the row direction as the first match error reference maximum value; obtaining the second match error reference maximum value comprises: obtaining a second number of blocks with successively increasing match errors based on the central block along the column direction as a vertical prediction radius; and obtaining the match error of the block corresponding to the vertical prediction radius at a distance from the central block along the column direction as the second match error reference maximum value.

3 . The scene detection method of claim 2 , wherein obtaining the horizontal prediction radius comprises:
obtaining a number of blocks with successively increasing match errors from the current central block along a first horizontal direction of the row as a first horizontal prediction number of blocks; obtaining a number of blocks with successively increasing match errors from the current central block along a second horizontal direction of the row as a second horizontal prediction number of blocks; and taking a smaller of the first horizontal prediction number of blocks and the second horizontal prediction number of blocks as the first number of blocks; obtaining the vertical prediction radius comprises: obtaining a number of blocks with successively increasing match errors from the current central block along a first vertical direction of the column as a first vertical prediction number of blocks; obtaining a number of blocks with successively increasing match errors from the current central block along a second vertical direction of the column as a second vertical prediction number of blocks; and taking a smaller of the first vertical prediction number of blocks and the second vertical prediction number of blocks as the second number of blocks.

4 . The scene detection method of claim 2 , wherein obtaining the first number of blocks with successively increasing match errors based on the central block along the row direction as the horizontal prediction radius comprises:
obtaining a first match error corresponding to a block located on one side of the central block along the row direction and at a distance from the central block with the horizontal prediction radius; obtaining a second match error corresponding to a block located on the other side of the central block along the row direction and at a distance from the central block with the horizontal prediction radius; and taking a smaller of the first match error and the second match error as the first match error reference maximum value; obtaining the match error of the block corresponding to the vertical prediction radius at a distance from the central block along the column direction as the second match error reference maximum value comprises: obtaining a third match error corresponding to a block located on one side of the central block along the column direction and at a distance from the central block with the vertical prediction radius; obtaining a fourth match error corresponding to a block located on the other side of the central block along the column direction and at a distance from the central block with the vertical prediction radius; and taking a smaller of the third match error and the fourth match error as the second match error reference maximum value.

5 . The scene detection method of claim 1 , wherein obtaining the first numerical comprises:
along the row direction of the search range, based on the relative magnitude of the match error of the block with the first horizontal threshold and with the second horizontal threshold, determining the number of times the match error of the corresponding block in each row changes periodically; and summing up the number of times the match error of the corresponding block for all row changes periodically to obtain the first numerical value; obtaining the second numerical comprises: along the column direction of the search range, based on the relative magnitude of the match error of the block with the first vertical threshold and with the second vertical threshold, determining the number of times the match error of the corresponding block in each column changes periodically; and summing up the number of times the match error of the corresponding block for all column changes periodically to obtain the second numerical value.

6 . The scene detection method of claim 5 , wherein determining the number of times the match error of the corresponding block in each row changes periodically comprises:
comparing the relative magnitude of the block match error with the first horizontal threshold in turn along the row to determine whether there is an incremental process of match error of the block; comparing the relative magnitude of the block match error with the second horizontal threshold in turn along the row to determine whether there is a decreasing process of match error of the block; based on the first horizontal threshold and the second horizontal threshold, marking a process of increasing match error and a process of decreasing match error as a period; and obtaining the number of the periods as a number of times the match error changes periodically for the corresponding block of the current row; determining the number of times the match error of the corresponding block in each column changes periodically comprises: comparing the relative magnitude of the block match error with the first vertical threshold in turn along the column to determine whether there is an increment process of match error of the block; comparing the relative magnitude of the block match error with the second vertical threshold in turn along the column to determine whether there is a decreasing process of match error of the block; based on the first vertical threshold and the second vertical threshold, marking a process of increas

Assignees

• Amlogic Shanghai Co Ltd

Inventors

• Bao Zheng
• Ji Tao
• He Zhihong
• Wang Chun
• Wang Dongjian
• Chen Xuyun

Classifications

• H04N5/145 Primary

  Movement estimation (for video coding H04N19/51) · CPC title

• H04N5/147

  Scene change detection · CPC title

• G06V10/993

  Evaluation of the quality of the acquired pattern · CPC title

• G06T2207/20021

  Dividing image into blocks, subimages or windows · CPC title

• G06T2207/20182

  Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering · CPC title