Unlock instant, AI-driven research and patent intelligence for your innovation.

Method and system for quickly retrieving photo for sketch map

A technology of sketching diagrams and photos, applied in the field of cross-modal analysis of image recognition, can solve problems such as indistinguishability, only focus on similarity between classes or changes within classes, ignore appearance differences, etc., to achieve the effect of improving accuracy

Pending Publication Date: 2021-11-30
NANJING UNIV OF POSTS & TELECOMM
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

(c) Inter-class similarity: similar people in photos and sketches will generate similar feature-level distributions, making it difficult to distinguish
(d) Intra-class diversity: Even if classes have the same label information, due to other external factors such as light, viewing angle and background, the distribution of features will also be uneven
To address the above challenges, existing methods either only focus on inter-class similarity or intra-class variation, or only focus on stable feature representations across different domains while ignoring appearance differences

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method and system for quickly retrieving photo for sketch map
  • Method and system for quickly retrieving photo for sketch map
  • Method and system for quickly retrieving photo for sketch map

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0036] Oriented sketch photograph fast retrieval method, comprising: collecting sketch; the input sketch sketch sketch retrieval model trained, the correlation output and input pictures of the condition is met.

[0037] like figure 1 As shown in the present embodiment for the sketch of the proposed embodiment flowchart of a method of quick photo retrieval is performed, which consists of four parts: 1) a network feature extraction; 2) against network; 3) triple network; 4) Sort network ; 5) correlation metric.

[0038] 1. The feature extraction network

[0039] Sketch retrieval model comprises a feature extraction network for extracting the image feature sketch input; DenseNet-169 employed in each branch as a shared feature extractor weight, referred to as F. (), Which contains four connected dense block convolution.

[0040] 2. confrontation Network

[0041] It includes against sketch retrieval model network for gradually reducing domain information and angle information, comprisi...

Embodiment 2

[0081] Based on the multi-sketch chart, a photo-rapid retrieval of the photographing, this embodiment provides a system for a quick-retrieval-to-selection map, including: a data acquisition module for collecting a sketch diagram; a sketch search module, Used to search the sketch map into a sketch search model, the output and the input sketch diagram meet the photos of the setting condition.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a method and a system for quickly retrieving photos for sketch maps in the technical field of cross-modal analysis of image recognition. The method for quickly retrieving photos for sketch maps comprises the following steps: acquiring a sketch map; and inputting the sketch map into a trained sketch map retrieval model, and outputting a photo of which relevancy with the input sketch map meets set conditions. According to the sketch map retrieval model, feature distribution of images from two fields is firstly aligned by using a domain adversarial sub-network; feature extraction interference caused by different shooting view angles is avoided through an angle adversarial sub-network, and triple loss and joint classification loss are designed for a triple network and a classification network and are respectively used for cross-modal features and reduction of distances between sketch-photo pairs; and finally, the relevancy between a specified sketch map and photos are calculated, and the photo with higher relevancy is the real photo of the sketch map. According to the invention, changes of fields and shooting angles can be solved on the feature level, and retrieval can be accurately carried out in a photo library according to sketch maps.

Description

Technical field [0001] Analysis Technical Field The present invention belongs to cross-modality image recognition, and particularly to a method and system for quick sketch picture retrieval proceed. Background technique [0002] Since the information carried by the sketch has a more intuitive features, making it an important way for people to express and record information. For tracking the target or specific event scene investigation, without being able to get photos, sketches an image that can only come through the information provided by witnesses goal by technical personnel. In this case, the sketch to exact match will have a photo of important significance, you can find and identify targets in the photo library by sketch drawn. Therefore, people photo retrieved quickly increasing attention to be oriented sketch. [0003] However, this task there are still several challenges: (a) inter-domain Gap: sketch is the lack of an abstract description of hue and texture information, a...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06F16/583G06F16/55G06K9/46
CPCG06F16/583G06F16/55
Inventor 鲍秉坤袁博闻陈柏茹
Owner NANJING UNIV OF POSTS & TELECOMM