Image feature binary coding representing method based on dot pair relation learning and reconstruction
A binary coding and image feature technology, applied in character and pattern recognition, special data processing applications, instruments, etc., can solve the problems of high feature dimension and low retrieval efficiency
Active Publication Date: 2018-09-14
NANJING UNIV
View PDF4 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0004] Purpose of the invention: The present invention aims at the problem of tight coupling between the hash function and the loss function of the optimization target in the existing unsupervised algorithm coding learning process, and proposes a loosely coupled image feature binary coding based on point pair relationship learning and reconstruction Representation method, so as to improve the performance and accuracy of image retrieval, effectively solve the problem of fast and accurate retrieval of images based on hash binary coded data
[0005] The image feature binary encoding representation method based on point-to-relationship learning and reconstruction constructed by the present invention aims to use machine learning and machine vision to reduce the high feature dimension and low retrieval efficiency in traditional image retrieval technology. In view of the problem of tight coupling between the hash function and the loss function of the optimization target in the encoding learning process of the existing unsupervised algorithm, the image retrieval performance can be improved by using the image feature binary encoding representation method based on point-to-relationship learning and reconstruction. and accuracy goals
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 moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment
[0145] This embodiment includes the following parts:
[0146] Import 10 data, each data is 784-dimensional image data. Figure 5 The pictures for 10 data are shown.
[0147] After the point-to-relation learning steps (including solving the coefficient matrix, dictionary segmentation, and feature representation), the reconstructed representation of the original imported data can be obtained, that is, the learning of the point-to-relationship is completed.
[0148] The reconstruction of these 10 data is expressed as follows, that is, the input data of the next step point pair relationship reconstruction step:
[0149] 0 0 0 0 0 1 0 0
[0150] 0 0 0 0 0 0 1 0
[0151] 0 1 0 0 0 0 0 0
[0152] 0 0 0 0 0 0 0 1
[0153] 0.0397 0 0 0 0.9603 0 0 0
[0154] 0.8445 0 0 0 0 0.1555 0 0
[0155] 0 0 1 0 0 0 0 0
[0156] 0 0 0 1 0 0 0 0
[0157] 0.8739 0 0 0 0 0 0.1261 0
[0158] 0 0.0534 0 0 0.9466 0 0 0
[0159] The next step is to reconstruct the point-to-point relationship, a...
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
Login to View More Abstract
The invention discloses an image feature binary coding representing method based on dot pair relation learning and reconstruction. The method comprises the following steps: step 1, converting data into a dictionary representation form, and solving a restricted problem, so as to obtain a coefficient matrix in the representation form; step 2, constructing a weight matrix of a representing graph to the coefficient matrix obtained in the step 1, and dividing a dictionary item on the graph into k groups, so as to realize dictionary segmentation; step 3, giving a new sample, computing the reconstruction residual of the new sample on a dictionary item group, and selecting an optimum dictionary item group corresponding to the minimum reconstruction residual to perform linear representation, so asto accomplish dot pair relation learning; step 4, solving an optimum model, learning optimum binary coding keeping the dot pair relation, and realizing dot pair relation reconstruction.
Description
technical field [0001] The invention belongs to the field of image feature encoding, in particular to a binary encoding representation method of image features based on point-to-relationship learning and reconstruction. Background technique [0002] In today's era, with the rapid development of the Internet information age, the total amount of image data is also increasing rapidly. In the application of image retrieval, given a query image, the user needs to retrieve images similar to it from a large-scale database, and return the results according to the similarity ranking. For this application scenario, one of the most basic methods is: firstly extract features from the query image and the database image respectively. Then, the distance between the query image and each database image is calculated according to a certain metric (such as Euclidean distance). Finally, the database images are sorted according to the distance, and the top database images are returned as the r...
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
Login to View More IPC IPC(8): G06F17/30G06K9/62
CPCG06F18/2323G06F18/28
Inventor杨育彬甘元柱毛晓蛟
OwnerNANJING UNIV