Covariant local feature aggregated image feature representation method
A local feature aggregation and image feature technology, which is applied in special data processing applications, instruments, electrical digital data processing, etc., can solve the problem of large storage space
Inactive Publication Date: 2016-02-17
XIAMEN UNIV
View PDF3 Cites 3 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
The advantage of the former is that it is easy to add various optimization schemes for retrieval results, such as adding various visual and geometric verifications, but the disadvantage is that it requires a large storage space
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 Construction
[0026] Embodiments of the present invention include the following steps:
[0027] 1) Extract the local features of the image and use the descriptors to describe. An image is represented as a set of descriptors χ, and the main direction θ of each local feature is obtained at the same time;
[0028] 2) Use an independent image set to extract and collect its local features to generate a matrix composed of descriptors; use the K-average clustering algorithm to obtain K cluster centers, and K is generally set between 32 and 64 An integer value, K cluster centers as the visual vocabulary set C = {c 1...K};
[0029] 3) Use the following formula (1) to perform simplified Fisher aggregation on each local feature of a picture:
[0030]
[0031] where q(x) finds the nearest neighbor c in C i ,b=B(θ);c i is the visual vocabulary set C = {c 1...K} in the visual vocabulary closest to the local feature x(x∈χ); the function B(θ) quantifies the main direction θ of the local feature x o...
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 provides a covariant local feature aggregated image feature representation method and relates to computer vision and multimedia information retrieval. The method comprises the steps of extracting the local features of images; offline training a small visual word set; aggregating the local feature of each image by using a simplified Fisher kernel method; during the aggregating, simultaneously considering the main direction information of each local feature, dividing the main direction into eight quantized intervals, aggregating to different Fisher sub vectors according to a quantized main direction value, splicing eight sub vectors into a long vector to be served as the feature representation of the images; recombining the eight Fisher sub vectors to obtain a series of 8d sub vectors; performing one-dimensional discrete cosine transform on each sub vector, and transforming to a frequency domain of the feature; then recombining the features of the frequency domain to obtain eight sub vectors of different frequency bands, performing a main component analysis on each sub vector, and recombining the eight sub vectors subjected to dimension reduction to obtain a series of 8d sub vectors; defining similarity measurement on the sub vectors to calculate the similarity of every two images.
Description
technical field [0001] The invention relates to computer vision and multimedia information retrieval, in particular to an image feature representation method of covariant local feature aggregation. Background technique [0002] With the introduction of Web2.0 and the popularity of various portable mobile multimedia devices, such as smart phones, iPads, digital cameras, etc., there are hundreds of billions of multimedia resources on the Internet. These multimedia resources mainly exist in the form of images and videos. According to statistics in 2014, more than 3,300 photos are uploaded to Flickr, the world's largest photo sharing website, every minute. The total number of photos maintained by Flickr has exceeded 7 billion. On Youku, the largest video sharing website in China, the total number of videos reached 45 million in 2010, with an average daily upload of 60,000. Note that the data scale of the above photo and video sharing websites is still growing at a relatively ...
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 Patent Type & Authority Applications(China)
IPC IPC(8): G06F17/30
CPCG06F16/5838
Inventor 赵万磊王菡子
Owner XIAMEN UNIV