US 2024/0185571 A1Published application

Image feature extraction method and apparatus, and computer device and readable storage medium

AssigneeSouthern University of Science and Technology

InventorsTao YE, Qi WANG, Jianghan ZHU, Shihang WANG, Xingbo WANG

Life of this patent

Estimating the term…

Loading patent…

Abstract

An image feature extraction method and apparatus are provided, along with a computer device and a readable storage medium. After obtaining the target convolutional kernel of the feature extraction convolutional neural network, this application will cut the input feature map obtained from the preprocessed image according to the preset image size into multiple abutting input feature submaps. Each input feature submap will be zero-padded to obtain the preprocessed feature submap. Then, for each preprocessed feature submap, the Karatsuba algorithm will be used to reduce the multiplier resource consumption during the convolution operation between the target convolutional kernel and the preprocessed feature submap, effectively suppressing the growth of adder resource consumption during the convolution operation, resulting in the output feature submap. Subsequently, multiple output feature submaps will be overlapped and the boundary of the overlapped feature map will be cropped to obtain the output feature map.

Claims13

Dependent claims

An image feature extraction method, comprising:

obtaining an original feature map for a target convolution kernel of a feature extraction convolution neural network;
performing image preprocessing on the original feature map according to a preset image size, so as to obtain a corresponding input feature map, wherein an image size of the input feature map is an integral multiple of the preset image size;
cutting the input feature map into a plurality of abutting input feature submaps according to the preset image size, wherein an image size of each of the input feature submaps is consistent with the preset image size;
for each of the input feature submaps, performing zero-padding on an input feature submap according to a convolution kernel size of the target convolution kernel, so as to obtain a corresponding preprocessed feature submap;
for each preprocessed feature submap, performing convolution operation on the preprocessed feature submap by using the target convolution kernel based on a Karatsuba algorithm, so as to obtain a corresponding output feature submap;
for a plurality of obtained output feature submaps, overlapping the output feature submaps with adjacent corresponding positions according to a distribution condition of each of the input feature submaps in the input feature map, so as to obtain an overlapped feature map;
and
cropping a boundary of the overlapped feature map according to the convolution kernel size of the target convolution kernel, so as to obtain a corresponding output feature map, wherein the output feature map is used to represent a feature extraction results obtained by computing the input feature map with its corresponding filter.

5 more elements

An image feature extraction apparatus, comprising:

a feature map obtaining module, configured to obtain an original feature map for a target convolution kernel of a feature extraction convolution neural network;
a feature map preprocessing module, configured to perform image preprocessing on the original feature map according to a preset image size, so as to obtain a corresponding input feature map, wherein an image size of the input feature map is an integral multiple of the preset image size;
an input feature cutting module, configured to cut the input feature map into a plurality of abutting input feature submaps according to the preset image size, wherein an image size of each of the input feature submaps is consistent with the preset image size;
a submap zero-padding module, configured to, for each of the input feature submaps, perform zero-padding on an input feature submap according to a convolution kernel size of the target convolution kernel, so as to obtain a corresponding preprocessed feature submap;
a feature convolution operation module, configured to, for each preprocessed feature submap, perform convolution operation on the preprocessed feature submap by using the target convolution kernel based on a Karatsuba algorithm, so as to obtain a corresponding output feature submap;
a submap overlapping module, configured to, for a plurality of obtained output feature submaps, overlap the output feature submaps with adjacent corresponding positions according to a distribution condition of each of the input feature submaps in the input feature map, so as to obtain an overlapped feature map;
and
an output feature extraction module, configured to crop a boundary of the overlapped feature map according to the convolution kernel size of the target convolution kernel, so as to obtain a corresponding output feature map, wherein the output feature map is used to represent a feature extraction results obtained by computing the input feature map with its corresponding filter.

5 more elements

Dependencies are inferred from the claim text.

Facts

Application: 18/396,324Filed 26 Dec 2023
Priority: CN 202110917990.511 Aug 2021
Published: 6 Jun 2024
Term ends: Not estimatedThe source records don’t support an estimate.

Classification6 codes

G06V 10/7715
Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or…:…
G06V 10/22
Image preprocessing: by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide…
G06V 10/82
Using pattern recognition or machine learning: using neural networks
Y02D 10/00
Energy efficient computing