CN104794210A - Image retrieval method combining visual saliency and phrases - Google Patents

Abstract

The invention discloses an image retrieval method combining visual saliency and phrases. The image retrieval method comprises the following steps that a search image is input and decoded into a YUV color space, and K-means clustering is conducted on pixels of the YUV color space, so that the image is divided into multiple superpixel units; likelihood computation is conducted on each superpixel unit of the divided image, and four obtained likelihoods are fused, and then the saliency map of the superpixel precision is obtained; bilateral Gaussian filtering is conducted, an image saliency map of the pixel precision is obtained; self-adaptive threshold segmentation is conducted on the image saliency map, and a binary image with a prominent target portion is obtained; a dictionary is established, vision words of an image salient region are extracted, and image description is generated; the image similarity between the search image and each image in a image base is calculated. By the adoption of the image retrieval method, the salient region in the image can be more accurately reflected, the visual saliency and the phases are well combined, and the retrieval effect is good.

Description

A kind of image search method in conjunction with vision significance and phrase

Technical field

Image processing field of the present invention, is specifically related to a kind of image search method in conjunction with vision significance and phrase.

Background technology

Developing rapidly and applying along with computing machine, network and multimedia technology, the quantity of digital picture increases just with surprising rapidity, and the image how to find people to need from mass digital image collection quickly and efficiently becomes a problem demanding prompt solution.For this reason, image retrieval technologies is arisen at the historic moment and is achieved very large development, from the earliest based on the retrieval that image manually marks, develops into the retrieval of present image content-based, precision and the efficiency of image retrieval are also all significantly increased, but still cannot meet the demand of people.The key of its problem is also do not have a kind of method can make the understanding image, semantic of computing machine completely as people at present.If the real meaning of image can be excavated further, and accurately expresses in a computer, the effect of image retrieval will certainly be promoted.

Summary of the invention

For solving the problem, the invention provides a kind of image search method in conjunction with vision significance and phrase.

For achieving the above object, the technical scheme that the present invention takes is:

In conjunction with an image search method for vision significance and phrase, comprise the steps:

Image decoding is YUV color space by S1, input inquiry image, and by carrying out K mean cluster to the pixel of YUV color space thus be some super-pixel unit by Iamge Segmentation;

S2, Likelihood Computation is carried out to each super-pixel unit of image after segmentation, obtain the measure value of this query image different parameters, Likelihood Computation comprise color independence is estimated, Color-spatial distribution is estimated, sports independence is estimated and the calculating of the space measure of spread;

S3, each for step S2 gained parameter estimated within linear normalization to [0,1] scope; Four kinds of fusions estimated are carried out to each super-pixel unit, obtains the Saliency maps of super-pixel precision;

S4, bilateral gaussian filtering is carried out to the Saliency maps of step S3 gained super-pixel precision, obtain the saliency figure of pixel precision;

S5, according to maximum variance between clusters, adaptive threshold fuzziness is carried out to the saliency figure of step S4 gained, obtain the binary map with outstanding significant target part;

S6, utilize SIFT algorithm to extract SIFT feature point in different classes of image from inquiry picture library, all unique point vector sets are incorporated into one piece, utilize K-Means clustering algorithm to merge similar SIFT feature point, construct the dictionary that comprises several vocabulary;

The visual word in S7, extraction saliency region, the number of visual word in statistical picture Saliency maps, structure visual phrase, the iamge description of synthetic image;

In S8, calculating query image and picture library, the image similarity of every width image, sorts to all images in picture library according to Similarity value, and returns associated picture on request as Query Result.

Wherein, the span of described judgment threshold is 0.15 ~ 0.35.

The present invention has following beneficial effect:

The salient region in query image can be reflected more exactly, vision significance is well combined with phrase, there is good retrieval effectiveness.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of a kind of image search method in conjunction with vision significance and phrase of the embodiment of the present invention.

Embodiment

In order to make objects and advantages of the present invention clearly understand, below in conjunction with embodiment, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

As shown in Figure 1, embodiments provide a kind of image search method in conjunction with vision significance and phrase, comprise the steps:

Image decoding is YUV color space by S1, input inquiry image, and by carrying out K mean cluster to the pixel of YUV color space thus query image being divided into some super-pixel unit;

S2, Likelihood Computation is carried out to each super-pixel unit of image after segmentation, obtain the measure value of this query image different parameters, Likelihood Computation comprise color independence is estimated, Color-spatial distribution is estimated, sports independence is estimated and the calculating of the space measure of spread;

S3, each for step S2 gained parameter estimated within linear normalization to [0,1] scope; Four kinds of fusions estimated are carried out to each super-pixel unit, obtains the Saliency maps of super-pixel precision;

S4, bilateral gaussian filtering is carried out to the Saliency maps of step S3 gained super-pixel precision, obtain the saliency figure of pixel precision;

S5, according to maximum variance between clusters, adaptive threshold fuzziness is carried out to the saliency figure of step S4 gained, obtain the binary map with outstanding significant target part;

S6, utilize SIFT algorithm to extract SIFT feature point in different classes of image from inquiry picture library, all unique point vector sets are incorporated into one piece, utilize K-Means clustering algorithm to merge similar SIFT feature point, construct the dictionary that comprises several vocabulary;

The visual word in S7, extraction saliency region, the number of visual word in statistical picture Saliency maps, structure visual phrase, the iamge description of synthetic image;

In S8, calculating query image and picture library, the image similarity of every width image, sorts to all images in picture library according to Similarity value, and returns associated picture on request as Query Result.

The span of described judgment threshold is 0.15 ~ 0.35.

The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (2)

1., in conjunction with an image search method for vision significance and phrase, it is characterized in that, comprise the steps:

Image decoding is YUV color space by S1, input inquiry image, and by carrying out K mean cluster to the pixel of YUV color space thus query image being divided into some super-pixel unit;

S2, Likelihood Computation is carried out to each super-pixel unit of image after segmentation, obtain the measure value of this query image different parameters, Likelihood Computation comprise color independence is estimated, Color-spatial distribution is estimated, sports independence is estimated and the calculating of the space measure of spread;

S3, each for step S2 gained parameter estimated within linear normalization to [0,1] scope; Four kinds of fusions estimated are carried out to each super-pixel unit, obtains the Saliency maps of super-pixel precision;

S4, bilateral gaussian filtering is carried out to the Saliency maps of step S3 gained super-pixel precision, obtain the saliency figure of pixel precision;

S5, according to maximum variance between clusters, adaptive threshold fuzziness is carried out to the saliency figure of step S4 gained, obtain the binary map with outstanding significant target part;

S6, utilize SIFT algorithm to extract SIFT feature point in different classes of image from inquiry picture library, all unique point vector sets are incorporated into one piece, utilize K-Means clustering algorithm to merge similar SIFT feature point, construct the dictionary that comprises several vocabulary;

The visual word in S7, extraction saliency region, the number of visual word in statistical picture Saliency maps, structure visual phrase, the iamge description of synthetic image;

In S8, calculating query image and picture library, the image similarity of every width image, sorts to all images in picture library according to Similarity value, and returns associated picture on request as Query Result.

2. a kind of image search method in conjunction with vision significance and phrase according to claim 1, it is characterized in that, the span of described judgment threshold is 0.15 ~ 0.35.