CN104142946A

CN104142946A - Method and system for aggregating and searching service objects of same type

Info

Publication number: CN104142946A
Application number: CN201310167405.XA
Authority: CN
Inventors: 邓宇; 欧海峰
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2013-05-08
Filing date: 2013-05-08
Publication date: 2014-11-12
Also published as: HK1201952A1

Abstract

The embodiment of the application provides a method for aggregating service objects of the same type. The method comprises the steps of: obtaining images of the service objects; generating digital signatures of the images based on visual characteristics of the images; establishing an image signature library by adopting the images and corresponding digital signatures; dividing images in the image signature library into multiple groups based on the digital signatures; calculating similarity of the images based on the multiple groups; clustering the images based on the similarity to form one or more image collections; combining the service objects corresponding to the images which belong to the same image collection into the service objects of the same type. The embodiment of the application is suitable for treatment of service object data in a large scale, and is high in aggregation efficiency.

Description

The method and system of the polymerization of a kind of same money business object, search

Technical field

The embodiment of the present application relates to the technical field of data search, particularly relates to the method for a kind of same money business object polymerization, the system of a kind of same money business object polymerization, a kind of searching method and a kind of search system.

Background technology

In recent years, along with the high speed development of internet, online quantity of information sharply increases, and has wherein comprised a large amount of image informations.Along with popularizing of various image handling implements, the processing such as the convergent-divergent of image, cutting, interpolation watermark all cause pro forma variation to image, but picture material is basic identical in fact.Excessive superimposed images information makes people will carry out loaded down with trivial details artificial filtration in information retrieval, expends time in and energy.

For example, in vertical search environment, when user carries out a certain business object (such as commodity) search by website, with money business object, be attribute (for example, the descriptions of the image of business object, the title of business object or business object etc.) complete or almost identical business object, in single Search Results, has very large probability to be demonstrated repeatedly.The main efficiency of information and the efficiency of user's obtaining information of providing in website has been provided in the displaying that repeats with money business object.

At present, with the polymerization of money business object, can utilize the modes such as type, parameter or title of business object to carry out.But when the loss of learnings such as type, parameter and title of business object or these information are not enough to carry out the differentiation of type of different money business objects, often need by manually screening, polymerization.Especially, because the type of the commodity of issuing in network environment and quantity are all in rapid growth, hand picking is more and more higher in time and human cost with the mode of money commodity, to such an extent as to cannot bear.

Therefore, need at present the urgent technical matters solving of those skilled in the art to be exactly: in information search engine, in the magnanimity merchandise news how to issue in internet, the same or analogous product of content is polymerized to the type with a business object, to improve the efficiency of information search.

Application content

The embodiment of the present application technical matters to be solved is to provide the polymerization of a kind of same money business object and a kind of searching method, can sign by the picture material generating digital based on business object, and can carry out fast verification, in order to realize in magnanimity commodity the polymerization with money business object, when guaranteeing polymerization effect, improve polymerization speed, this is applied to search field, improve the efficiency of information search.

Accordingly, the embodiment of the present application also provides system and a kind of search system of a kind of same money business object polymerization, in order to guarantee the implementation and application of said method.

In order to address the above problem, the application discloses the method for a kind of same money business object polymerization, comprising:

Obtain the image of business object;

According to the visual signature of described image, generate the digital signature of described image;

Adopt described image and corresponding digital signature thereof to set up image signatures storehouse;

According to described digital signature, the image in image signatures storehouse is divided into a plurality of groupings;

Based on described a plurality of groupings, calculate the similarity of described image;

According to described similarity, described image is carried out to cluster, form one or more image collections;

Business object corresponding to the image that belongs to same image collection merged into same money business object.

Preferably, the step that the described visual signature according to image generates the digital signature of described image comprises:

Extract respectively the visual signature of described image; Described visual signature comprises color characteristic, and/or, Gradient Features, and/or, local feature;

Described visual signature is normalized, is combined as the digital signature of correspondence image.

Preferably, described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, the described step of extracting respectively the visual signature of image comprises:

The color dimension that acquisition quantizes in default color space;

Travel through the color value of each pixel in described image, search the color dimension with the color value ownership of described each pixel; Wherein, the color dimension of the color value of described each pixel ownership is the color dimension nearest with the color value of each pixel;

Add up each color dimension corresponding pixel number in described image, select color dimension that one or more pixel numbers are maximum as the main color of present image;

Obtain the pixel number of described main color correspondence in present image;

And/or,

When described visual signature comprises Gradient Features, the described step of extracting respectively the visual signature of image comprises:

Described image is converted into gray level image, and described gray level image is carried out smoothly;

According to described gray level image after level and smooth, calculate the gradient orientation histogram of described image;

Adopt the pixel number of described gradient orientation histogram and correspondence thereof as the Gradient Features of present image;

And/or,

When described visual signature comprises local feature, the described step of extracting respectively the visual signature of image comprises:

Extract the unique point in described image with rotational invariance and yardstick unchangeability;

Calculate the contrast variable of described unique point, choose the local feature that unique point that contrast variable is greater than default the first threshold value is spliced into described image.

Preferably, the sub-step that the gray level image of described foundation after level and smooth calculates the gradient orientation histogram of described image further comprises:

For the gray level image after level and smooth, calculate gradient direction and the gradient magnitude of each pixel;

Present image is done to gradient direction statistics, generate the histogram that gradient direction is transverse axis of take of present image;

The gradient direction of described present image is divided into R orientation angle, generates R R set of histograms distance corresponding to orientation angle difference; Wherein, described R is positive integer;

All pixels in traversing graph picture, find out set of histograms distance corresponding to immediate both direction angle according to the gradient direction of each pixel respectively, and according to the degree of closeness coefficient that assigns weight from high to low;

The gradient magnitude of described pixel is multiplied by weight coefficient to be added to respectively in set of histograms distance corresponding to described immediate both direction angle;

Be normalized, generate the gradient orientation histogram of present image.

Preferably, the described step that image in image signatures storehouse is divided into a plurality of groupings according to digital signature comprises:

Extract the color characteristic in the digital signature of described image;

From described color characteristic, extract the main color of grouping; Wherein, the main color of described grouping comprises the main color that maximum pixel numbers are corresponding; And/or, by calculating in described image the pixel quantity of main color corresponding to pixel numbers at most, account for the ratio of all pixel quantity of described image, when described ratio is greater than the second predetermined threshold value, selected corresponding pixel points many for several times main colors;

According to the main color of described grouping, described image is divided into a plurality of groupings, in same grouping, the main color of the grouping of image is identical.

Preferably, the described step of calculating the similarity of described image based on a plurality of groupings comprises:

Determine every main grouping and adjacent packets that image is corresponding, described master is grouped into the grouping at present image place, and described adjacent packets is the grouping the highest with described main grouping similarity; Described adjacent packets is one or more;

For every image, calculate the similarity of the color characteristic of other image in main grouping that its color characteristic is corresponding with it and adjacent packets;

The similarity of removing described color characteristic in described main grouping and adjacent packets is less than the image of the 3rd predetermined threshold value;

For present image, calculate the similarity of the Gradient Features of residual image in main grouping that its Gradient Features is corresponding with it and adjacent packets;

The similarity of further removing described Gradient Features in described main grouping and adjacent packets is less than the image of the 4th predetermined threshold value;

For present image, calculate the similarity of the local feature of last remaining image in main grouping that its local feature is corresponding with it and adjacent packets;

Similarity using the similarity of the local feature of last remaining image in the main grouping corresponding with it of described present image and adjacent packets as described image.

Preferably, describedly according to described similarity, described image is carried out to cluster, the step that forms one or more image collections comprises:

If the similarity of described image, higher than the 5th predetermined threshold value, is put into same image collection by described image.

Preferably, described method also comprises:

The image that simultaneously appears at the same business object in a plurality of image collections is carried out to duplicate removal processing.

Preferably, the described step that the image that simultaneously appears at the same business object in a plurality of image collections is carried out to duplicate removal processing comprises:

Set up the image tree corresponding with each image collection, the root node of described image tree is set to the image of current business object, and the leaf node of described image tree is set to all images of described image collection;

Travel through all leaf nodes of described image tree, the image tree that all leaf nodes are corresponding merges in present image tree, eliminates merged image tree;

Travel through all image trees, remove the image of the business object of all repetitions.

The embodiment of the present application discloses a kind of searching method, comprising:

Receive user's searching request;

According to described request, obtain Search Results; And

The business object in described Search Results with identical class indication is merged into same money business object;

Wherein, the generation method of described class indication comprises:

Obtain the image of business object;

Adopt described image and corresponding digital signature thereof to set up image signatures storehouse; According to described digital signature, the image in image signatures storehouse is divided into a plurality of groupings;

Similarity based on described a plurality of grouping computed image;

Business object corresponding to the image that belongs to same image collection distributed to same class indication.

The color dimension that acquisition quantizes in default color space;

And/or,

Preferably, the generation method of described class indication also comprises:

The embodiment of the present application discloses the system of a kind of same money business object polymerization, comprising:

Image collection module, for obtaining the image of business object;

Digital signature generation module, for generating the digital signature of described image according to the visual signature of described image;

Module is set up in image signatures storehouse, for adopting described image and corresponding digital signature thereof to set up image signatures storehouse;

Module is divided in grouping, for the image in image signatures storehouse being divided into a plurality of groupings according to described digital signature;

Similarity calculation module, for calculating the similarity of described image based on described a plurality of groupings;

Image collection forms module, for described image being carried out to cluster according to described similarity, forms one or more image collections;

With money business object, merge module, for business object corresponding to the image that belongs to same image collection merged into same money business object.

Preferably, described digital signature generation module comprises:

Visual Feature Retrieval Process submodule, for extracting respectively the visual signature of described image; Described visual signature comprises color characteristic, and/or, Gradient Features, and/or, local feature;

Digital signature combination submodule, for described visual signature is normalized, is combined as the digital signature of correspondence image.

Preferably, described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, described Visual Feature Retrieval Process submodule comprises:

Color dimension obtains submodule, for obtaining the color dimension quantizing at default color space;

Pixel ownership is searched submodule, for traveling through the color value of described each pixel of image, searches the color dimension with the color value ownership of described each pixel; Wherein, the color dimension of the color value of described each pixel ownership is the color dimension nearest with the color value of each pixel;

Main color generates submodule, for adding up each color dimension in pixel number corresponding to described image, selects color dimension that one or more pixel numbers are maximum as the main color of present image;

Pixel number is obtained submodule, for obtaining described main color in pixel number corresponding to present image;

And/or,

When described visual signature comprises Gradient Features, described Visual Feature Retrieval Process submodule comprises:

Greyscale image transitions submodule, for described image is converted into gray level image, and carries out smoothly described gray level image;

Gradient orientation histogram calculating sub module, for calculating the gradient orientation histogram of described image according to described gray level image after level and smooth;

Gradient Features generates submodule, for adopting the pixel number of described gradient orientation histogram and correspondence thereof as the Gradient Features of present image;

And/or,

When described visual signature comprises local feature, described Visual Feature Retrieval Process submodule comprises:

Feature point extraction submodule, has the unique point of rotational invariance and yardstick unchangeability for extracting described image;

Local feature generates submodule, for calculating the contrast variable of described unique point, chooses the local feature that unique point that contrast variable is greater than default the first threshold value is spliced into described image.

Preferably, described gradient orientation histogram calculating sub module further comprises:

Gradient direction and gradient magnitude calculating sub module, for the gray level image for after level and smooth, calculate gradient direction and the gradient magnitude of each pixel;

The first histogram generates submodule, for present image being done to gradient direction statistics, generates the histogram that gradient direction is transverse axis of take of present image;

The second histogram generates submodule, for the gradient direction of described present image being divided into R orientation angle, generates R R set of histograms distance corresponding to orientation angle difference; Wherein, described R is positive integer;

Weight allocation submodule, for all pixels of traversing graph picture, finds out set of histograms distance corresponding to immediate both direction angle according to the gradient direction of each pixel respectively, and according to the degree of closeness coefficient that assigns weight from high to low;

The weight submodule that adds up, for the gradient magnitude of described pixel is multiplied by weight coefficient be added to respectively set of histograms corresponding to described immediate both direction angle apart from;

Gradient orientation histogram generates submodule, for being normalized, generates the gradient orientation histogram of present image.

Preferably, described grouping division module comprises:

Color characteristic extracts submodule, for extracting the color characteristic of the digital signature of described image;

Main color is chosen submodule, for extract the main color of grouping from described color characteristic; Wherein, the main color of described grouping comprises the main color that maximum pixel numbers are corresponding; And/or, by calculating in described image the pixel quantity of main color corresponding to pixel numbers at most, account for the ratio of all pixel quantity of described image, when described ratio is greater than the second predetermined threshold value, selected corresponding pixel points many for several times main colors;

Image packets submodule, for described image being divided into a plurality of groupings according to the main color of described grouping, in same grouping, the main color of the grouping of image is identical.

Preferably, described similarity calculation module comprises:

Submodule is determined in grouping, and for determining every main grouping and adjacent packets that image is corresponding, described master is grouped into the grouping at present image place, and described adjacent packets is the grouping the highest with described main grouping similarity; Described adjacent packets is one or more;

The first similarity calculating sub module, for for every image, calculates the similarity of the color characteristic of other image in main grouping that its color characteristic is corresponding with it and adjacent packets;

The first image is removed submodule, is less than the image of the 3rd predetermined threshold value for remove the similarity of described color characteristic in described main grouping and adjacent packets;

The second similarity calculating sub module, for for present image, calculates the similarity of the Gradient Features of residual image in main grouping that its Gradient Features is corresponding with it and adjacent packets;

The second image is removed submodule, is less than the image of the 4th predetermined threshold value for the similarity at described main grouping and the described Gradient Features of the further removal of adjacent packets;

Third phase is seemingly spent calculating sub module, for for present image, calculates the similarity of the local feature of last remaining image in main grouping that its local feature is corresponding with it and adjacent packets;

Similarity is determined submodule, for the similarity using the similarity of the local feature of the main grouping corresponding with it of described present image and the last remaining image of adjacent packets as described image.

Preferably, described image collection formation module comprises:

Image is put into submodule, for the similarity when described image, during higher than the 5th predetermined threshold value, described image is put into same image collection.

Preferably, described system also comprises:

Duplicate removal module, for carrying out duplicate removal processing by the image that appears at the same business object of a plurality of image collections simultaneously.

Preferably, described duplicate removal module comprises:

Image tree is set up submodule, and for setting up the image tree corresponding with each image collection, the root node of described image tree is set to the image of current business object, and the leaf node of described image tree is set to all images of described image collection;

Image tree merges submodule, and for traveling through all leaf nodes of described image tree, the image tree that all leaf nodes are corresponding merges in present image tree, eliminates merged image tree;

Image tree traversal submodule, for traveling through all image trees, removes the image of the business object of all repetitions.

The embodiment of the present application discloses a kind of search system, comprising:

Request receiving module, for receiving user's searching request;

Search Results acquisition module, for according to described request, obtains Search Results; And

Search Results merges module, for described Search Results being had to the business object of identical class indication, merges into same money business object;

Wherein, described class indication generates by following submodule:

Image Acquisition submodule, for obtaining the image of business object;

Digital signature generates submodule, for generate the digital signature of described image according to the visual signature of described image;

Submodule is set up in image signatures storehouse, for adopting described image and corresponding digital signature thereof to set up image signatures storehouse;

Submodule is divided in grouping, for the image in image signatures storehouse being divided into a plurality of groupings according to described digital signature;

Similarity calculating sub module, for calculating the similarity of described image based on described a plurality of groupings;

Image collection forms submodule, for described image being carried out to cluster according to described similarity, forms one or more image collections;

With money business object, merge submodule, for business object corresponding to the image that belongs to same image collection merged into same money business object;

Class indication distribution sub module, for distributing same class indication by business object corresponding to the image that belongs to same image collection.

Preferably, described digital signature generation submodule comprises:

And/or,

Preferably, described Search Results merging module also comprises:

Duplicate removal submodule, for carrying out duplicate removal processing by the image that appears at the same business object of a plurality of image collections simultaneously.

Compare with background technology, the embodiment of the present application comprises following advantage:

The embodiment of the present application has proposed the scheme that picture material based on business object generates corresponding digital signature, this scheme is by extracting some the significant visual signatures in image, form the information combination of certain complexity, complete the abstractdesription to picture material, generate the identification mark of representative this image of conduct; Simultaneously, according to extracted visual signature, formulate corresponding proof rule and differentiate the multiplicity between different signatures, recognition correct rate is high, matching degree according to signature, condenses together the same or analogous image of content, and further will condense together with money business object, autopolymerization, feasibility is high.

The embodiment of the present application has also proposed a kind of level polymerization.First, the image for the treatment of polymerization according to Image Visual Feature divides into groups, and while carrying out images match, for an image, only the image in current group and in adjacent packets can be compared, and greatly reduces the number of times of images match; Secondly, while carrying out images match, level reasonable in design is processed structure, by the size of computational complexity, easy first and difficult later, first investigate relatively simple color, then being the medium Gradient Features of complexity, is finally more complicated local feature, successively filters the candidate image that the degree of correlation is low, the final same or analogous image of precise positioning content, polymerization generates the image collection with money business object.The embodiment of the present application operand is little, is applicable to the processing of extensive business object data, and polymerization efficiency is high.

The image collection of all same money business objects is carried out to polymerization again, according to the correlativity between the image collection of same money business object, generate the image polymerization result of final same money business object, remove the image of the business object repeating, reduce data redundancy.

Accompanying drawing explanation

Fig. 1 shows the flow chart of steps of embodiment of the method 1 of the application's a kind of same money business object polymerization;

Fig. 2 shows the flow chart of steps of embodiment of the method 2 of the application's a kind of same money business object polymerization;

Fig. 3 shows the flow chart of steps of a kind of searching method embodiment of the application;

Fig. 4 shows the structured flowchart of system embodiment of the application's a kind of same money business object polymerization;

Fig. 5 shows the structured flowchart of a kind of search system embodiment of the application.

Embodiment

For the application's above-mentioned purpose, feature and advantage can be become apparent more, below in conjunction with the drawings and specific embodiments, the application is described in further detail.

One of core idea of the embodiment of the present application is, proposes a kind of digital signature and corresponding proof rule that reflects picture material, so that when searching for, can carry out based on this digital signature the polymerization fast and accurately of same money business object.

With reference to Fig. 1, show the flow chart of steps of embodiment of the method 1 of the application's a kind of same money business object polymerization, specifically can comprise the steps:

Step 101, obtains the image of business object;

It should be noted that, the business object in the embodiment of the present application can comprise the concrete things in different business field.

For making those skilled in the art understand better the embodiment of the present application, in this manual, a kind of example using commodity as business object describes.

In specific implementation, commodity in the embodiment of the present application can be by main the shown a or many moneys commodity in one or more websites, in the information of described commodity, can comprise one or more item property, such as parameter of commodity image, trade name, commodity price, descriptive labelling, marque or commodity etc.So-called with money commodity, can be by same or the complete or almost identical commodity of main the shown item property of different web sites.

Step 102, generates the digital signature of described image according to the visual signature of described image;

In the application's a preferred embodiment, described step 102 specifically can comprise following sub-step:

Sub-step S11, extracts respectively the visual signature of described image; Described visual signature can comprise color characteristic, and/or, Gradient Features, and/or, local feature;

Sub-step S12, is normalized described visual signature, is combined as the digital signature of correspondence image.

Being appreciated that so-called visual signature, can be to have the features such as the shape of image of meaning directly perceived and color, the content that the corresponding digital signature generating can token image.

In a preferred exemplary of the embodiment of the present application, described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, described sub-step S11 specifically can comprise following sub-step:

Sub-step S11-11, obtains the color dimension quantizing in default color space;

For example, in RGB RGB color space, a pixel has red R, green G, tri-Color Channels of blue B, and the storage format of every kind of color is all 8 two numbers processed, be that each Color Channel just has the brightness of 2^8=256 kind, a pixel just has 256*256*256 kind color dimension.Because dimension is too many, in practical application, be difficult to statistical study, practicality is low, therefore need to reduce color dimension.Originally the brightness stepping of default color passage is 1, and supposing now to select brightness stepping is 16, and each Color Channel just has 16 kinds of brightness, and a pixel also just has 16*16*16 kind color dimension.

Sub-step S11-12, travels through the color value of each pixel in described image, searches the color dimension with the color value ownership of described each pixel; Wherein, the color dimension of the color value of described each pixel ownership is the color dimension nearest with the color value of each pixel;

It should be noted that described distance can include but not limited to Euclidean distance.If the color value of an image slices vegetarian refreshments is (130,234,111), the Euclidean distance d=sqrt ((130-R) ^2+ (234-G) ^2+ (111-B) ^2) of color dimension after itself and uniform quantization, can obtain the color dimension nearest with it by calculating so.

Sub-step S11-13, adds up each color dimension corresponding pixel number in described image, selects color dimension that one or more pixel numbers are maximum as the main color of present image;

In a preferred exemplary of the present embodiment, choose color dimension that 6 pixel numbers are maximum as the main color of present image.

Sub-step S11-14, obtains described main color corresponding pixel number in present image.

In specific implementation, the method that generates color characteristic according to the pixel number of described main color and correspondence thereof is not unique, and the embodiment of the present application is not limited this.

Described color characteristic can be described the color distribution of present image.Because color characteristic is mainly the effect of playing a scalping, thus must be quick as far as possible in the processing procedure of business object image, only need the general information of token image content.

In the application's a preferred embodiment, when described visual signature comprises Gradient Features, described sub-step S11 specifically can also comprise following sub-step:

Sub-step S11-21, is converted into gray level image by described image, and described gray level image is carried out smoothly;

In specific implementation, can remove the noise effect of image; Described carrying out smoothly, can include but not limited to use Gaussian smoothing to carry out smoothing processing.

Sub-step S11-22, calculates the gradient orientation histogram of described image according to described gray level image after level and smooth;

In a preferred exemplary of the present embodiment, described sub-step S11-22 further can comprise following sub-step:

Sub-step C1, for the gray level image after level and smooth, calculates gradient direction and the gradient magnitude of each pixel;

In specific implementation, can include but not limited to the gradient direction and the gradient magnitude that use first order difference to calculate each pixel.

Sub-step C2, does gradient direction statistics to present image, generates the histogram that gradient direction is transverse axis of take of present image;

In specific implementation, gradient direction can be got 0 ° to 180 °, also can get 0 ° to 360 °.

Sub-step C3, all pixels in traversing graph picture, find out set of histograms distance corresponding to immediate both direction angle according to the gradient direction of each pixel respectively, and according to the degree of closeness coefficient that assigns weight from high to low;

Sub-step C4, all pixels in traversing graph picture, find out histogram bin corresponding to immediate both direction angle according to the gradient direction of each pixel respectively, and according to the degree of closeness coefficient that assigns weight from high to low;

In specific implementation, described two weight coefficients and can be 1.

Sub-step C5, the gradient magnitude of just described pixel is multiplied by weight coefficient and is added to respectively in set of histograms distance corresponding to described immediate both direction angle;

In specific implementation, sub-step C4 and C5 are for the boundary effect between fuzzy different histogram bin.

Sub-step C6, is normalized, and generates the gradient orientation histogram of present image.

In specific implementation, all pixels in present image are carried out, after C4 and C5 operation, carrying out sub-step C6.

Sub-step S11-23, adopts the pixel number of described gradient orientation histogram and correspondence thereof as the Gradient Features of present image;

In specific implementation, the method for described generation Gradient Features is not unique, and the embodiment of the present application is not limited this.

In the application's a preferred embodiment, when described visual signature comprises local feature, described sub-step S11 specifically can also comprise following sub-step:

Sub-step S11-31, extracts the unique point in described image with rotational invariance and yardstick unchangeability;

In specific implementation, can adopt yardstick invariant features conversion SIFT as local description, to carry out the local feature of Description Image.It should be noted that, to those skilled in the art, the unique point (being key point) of utilizing SIFT method to calculate to have rotational invariance and yardstick unchangeability in image, belongs to known technology, and the embodiment of the present application does not elaborate at this.The quantity of SIFT unique point can be set according to actual conditions by those skilled in the art, and the embodiment of the present application is not limited this.

In a preferred exemplary of the present embodiment, can be centered by each key point, choose 64 pixels of surrounding, every 4 adjacent pixels are a pixel groups, obtain altogether 16 pixel groups, each pixel groups relatively obtains 8 coordinate figures with key point again, obtains altogether 128 coordinate figures, is the SIFT feature of this key point.SIFT feature is a kind of similar invariant in theory, all insensitive to convergent-divergent, displacement, rotation, makes SIFT feature have stronger adaptability to many variations of image.

Feature extraction can be divided into detection and describe two stages, take SIFT as example, and SIFT unique point is the result of feature detection, and the feature of above-mentioned 128 coordinate figures is the description of its corresponding SIFT unique point.

Sub-step S11-32, calculates the contrast variable of described unique point, chooses the local feature that unique point that contrast variable is greater than default the first threshold value is spliced into described image.

In a preferred exemplary of the present embodiment, can include but not limited to take pruning method to select SIFT unique point.

In another preferred exemplary of the present embodiment, described the first predetermined threshold value is 100.

Local feature antijamming capability is strong, can be for accurate coupling.

In specific implementation, the described method according to unique point splicing generation local feature is not unique, and the embodiment of the present application is not limited this.

Certainly, the method for said extracted Image Visual Feature, just as example, when implementing the embodiment of the present application, can arrange the method that other extract Image Visual Feature according to actual conditions, and the embodiment of the present application is not limited this.In addition, except the method for said extracted Image Visual Feature, those skilled in the art can also adopt other to extract the method for Image Visual Feature by actual needs, and the embodiment of the present application is not limited this yet.

The mode that adopts Image Visual Feature to generate corresponding digital signature can be set according to actual conditions by those skilled in the art, it can be a character string, by one or more the combining in color characteristic, Gradient Features and local feature, a corresponding character string of image, can store digital signature by the mode of binary file, the embodiment of the present application is not limited this yet.

Step 103, adopts described image and corresponding digital signature thereof to set up image signatures storehouse;

In specific implementation, the embodiment of the present application can be regularly or not timing obtain new merchandise news, based on image corresponding to commodity, generate corresponding digital signature, add in described image signatures storehouse, the embodiment of the present application is not limited this.

Step 104, is divided into a plurality of groupings according to described digital signature by the image in image signatures storehouse;

It should be noted that, the image dividing into groups can be all images in image signatures storehouse, can be also the parts of images in image signatures storehouse, and the embodiment of the present application is not limited this.

In the application's a preferred embodiment, described step 104 specifically can comprise following sub-step:

Sub-step S21, extracts the color characteristic in the digital signature of described image;

It should be noted that, the method for the digital signature of analysis diagram picture can be corresponding with the method for digital signature that generates business object image.

Sub-step S22 extracts the main color of grouping from described color characteristic; Wherein, the main color of described grouping comprises the main color that maximum pixel numbers are corresponding; And/or, by calculating in described image the pixel quantity of main color corresponding to pixel numbers at most, account for the ratio of all pixel quantity of described image, when described ratio is greater than the second predetermined threshold value, selected corresponding pixel points many for several times main colors;

It should be noted that, how inferior can be to be only second at most.

In a preferred exemplary of the present embodiment, described the second predetermined threshold value is the arbitrary value in 0.6～0.8.

Sub-step S23, is divided into a plurality of groupings according to the main color of described grouping by described image, and in same grouping, the main color of the grouping of image is identical.

Image is divided into groups, and is to image polymerization preliminary screening, can greatly reduce the data processing amount in image polymerization process.

Step 105, calculates the similarity of described image based on described a plurality of groupings;

In the application's a preferred embodiment, described step 105 specifically can comprise following sub-step:

Sub-step S31, determines every main grouping and adjacent packets that image is corresponding, and described master is grouped into the grouping at present image place, and described adjacent packets is the grouping the highest with described main grouping similarity; Described adjacent packets is one or more;

Be appreciated that the relation that can exist sequence between grouping, grouping the closer to, the similarity of the main color of dividing into groups can be higher, and the main color similarity of grouping between adjacent packets can be the highest.

Sub-step S32, for every image, calculates the similarity of the color characteristic of other image in main grouping that its color characteristic is corresponding with it and adjacent packets;

Being appreciated that the wherein similarity of an image and other images of calculating, can be to calculate the similarity of other images of described image and its place dividing into groups and the similarity of the image of described image and other groupings.

In a preferred exemplary of the present embodiment, select an adjacent packets to carry out the calculating of image similarity.

Sub-step S33, the similarity of removing described color characteristic in described main grouping and adjacent packets is less than the image of the 3rd predetermined threshold value;

Removal can represent between described image and not have similarity.

In a preferred exemplary of the present embodiment, described the 3rd predetermined threshold value is 0.875.

Sub-step S34, for present image, calculates the similarity of the Gradient Features of residual image in main grouping that its Gradient Features is corresponding with it and adjacent packets;

Sub-step S35, the similarity of further removing described Gradient Features in described main grouping and adjacent packets is less than the image of the 4th predetermined threshold value;

In a preferred exemplary of the present embodiment, described the 4th predetermined threshold value is 0.925.

Sub-step S36, for present image, calculates the similarity of the local feature of last remaining image in main grouping that its local feature is corresponding with it and adjacent packets;

Sub-step S37, the similarity using the similarity of the local feature of last remaining image in the main grouping corresponding with it of described present image and adjacent packets as described image.

Step 106, carries out cluster according to described similarity to described image, forms one or more image collections;

In the application's a preferred embodiment, described step 106 specifically can comprise following sub-step:

Sub-step S41, if the similarity of described image is higher than the 5th predetermined threshold value, puts into same image collection by described image.

In a preferred exemplary of the present embodiment, described the 5th predetermined threshold value is 0.95.

Result after polymerization can be that each image all can obtain a corresponding image collection.Essence is identical in terms of content with present image for the image of described image collection the inside.

Step 107, merges into same money business object by business object corresponding to the image that belongs to same image collection.

In the embodiment of the present application, commodity corresponding to the image of same image collection can be merged into same money commodity.

With reference to Fig. 2, show the flow chart of steps of embodiment of the method 2 of the application's a kind of same money business object polymerization, specifically can comprise the steps:

Step 201, obtains the image of business object;

Step 202, generates the digital signature of described image according to the visual signature of described image;

Step 203, adopts described image and corresponding digital signature thereof to set up image signatures storehouse;

Step 204, is divided into a plurality of groupings according to described digital signature by the image in image signatures storehouse;

Step 205, calculates the similarity of described image based on described a plurality of groupings;

Step 206, carries out cluster according to described similarity to described image, forms one or more image collections;

Step 207, carries out duplicate removal processing by the image that appears at the same business object in a plurality of image collections simultaneously;

Step 208, merges into same money business object by business object corresponding to the image that belongs to same image collection.

In the embodiment of the present application, described commodity, after polymerization, for every commodity image, can obtain an image collection corresponding with described image, and the image in described set is the same on content.But in polymerization result, unavoidably there will be the image of same commodity to appear in a plurality of image collections, in order to reduce data redundancy, need to remove the processing that repeats commodity image to all image collections.

In the application's a preferred embodiment, described step 207 specifically can comprise following sub-step:

Sub-step S51, sets up the image tree corresponding with each image collection, and the root node of described image tree is set to the image of current business object, and the leaf node of described image tree is set to all images of described image collection;

Sub-step S52, travels through all leaf nodes that described image is set, and the image tree that all leaf nodes are corresponding merges in present image tree, eliminates merged image tree;

Sub-step S53, travels through all image trees, removes the image of the business object of all repetitions.

For example, the image collection of commodity image M 0 comprises image M 1, M2 and M3, set up respectively image tree T0, T1, T2, T3 that M0, M1, M2, M3 are corresponding, wherein the root node of T0, T1, T2, T3 is respectively M0, M1, M2, M3, and the leaf node of T0 is respectively M1, M2 and M3.The leaf node of traversal T0, image corresponding to M1, M2 and M3 tree T1, T2 and T3 merge to T0, and T1, T2, all leaf nodes below T3 root node are connected to T0, and structure remains unchanged, and T1, T2, T3 are deleted.

In the embodiment of the present application, all image trees are carried out to aforesaid operations, until all images tree is all disposed.Complete and remove each image tree of repeating to be left after business object image is processed, the image collection of corresponding final same money commodity, on content, the set of identical image is only surplus next, and the set of the same money commodity of correspondence is also only surplus next.

Referring to Fig. 3, show the flow chart of steps of a kind of searching method embodiment of the application, specifically can comprise the steps:

Step 301, reception user's searching request;

Step 302, according to described request, obtains Search Results; And

Step 303, merges into same money business object by the business object in described Search Results with identical class indication;

Wherein, the generation method of described class indication specifically can comprise following sub-step:

Sub-step S61, obtains the image of business object;

Sub-step S62, generates the digital signature of described image according to the visual signature of described image;

Sub-step S63, adopts described image and corresponding digital signature thereof to set up image signatures storehouse; According to described digital signature, the image in image signatures storehouse is divided into a plurality of groupings;

Sub-step S64, the similarity based on described a plurality of grouping computed image;

Sub-step S65, carries out cluster according to described similarity to described image, forms one or more image collections;

Sub-step S66, distributes same class indication by business object corresponding to the image that belongs to same image collection.

In the application's a preferred embodiment, described sub-step S62 further can comprise following sub-step:

Sub-step S62-1, extracts respectively the visual signature of described image; Described visual signature comprises color characteristic, and/or, Gradient Features, and/or, local feature;

Sub-step S62-2, is normalized described visual signature, is combined as the digital signature of correspondence image.

In a preferred exemplary of the embodiment of the present application, described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, described sub-step S62-1 specifically can comprise following sub-step:

Sub-step D1, obtains the color dimension quantizing in default color space;

Sub-step D2, travels through the color value of each pixel in described image, searches the color dimension with the color value ownership of described each pixel; Wherein, the color dimension of the color value of described each pixel ownership is the color dimension nearest with the color value of each pixel;

Sub-step D3, adds up each color dimension corresponding pixel number in described image, selects color dimension that one or more pixel numbers are maximum as the main color of present image;

Sub-step D4, obtains described main color corresponding pixel number in present image;

In a preferred exemplary of the embodiment of the present application, when described visual signature comprises Gradient Features, described sub-step S62-1 specifically can also comprise following sub-step:

Sub-step E1, is converted into gray level image by described image, and described gray level image is carried out smoothly;

Sub-step E2, calculates the gradient orientation histogram of described image according to described gray level image after level and smooth;

Sub-step E3, adopts the pixel number of described gradient orientation histogram and correspondence thereof as the Gradient Features of present image;

In a preferred exemplary of the embodiment of the present application, when described visual signature comprises local feature, described sub-step S62-1 specifically can also comprise following sub-step:

Sub-step F1, extracts the unique point in described image with rotational invariance and yardstick unchangeability;

Sub-step F2, calculates the contrast variable of described unique point, chooses the local feature that unique point that contrast variable is greater than default the first threshold value is spliced into described image.

In the application's a preferred embodiment, the generation method of described class indication specifically can also comprise following sub-step:

Sub-step G1, carries out duplicate removal processing by the image that appears at the same business object in a plurality of image collections simultaneously.

In specific implementation, according to the request of user search commodity, search for, can obtain a traditional commercial articles searching result, with money commodity, may be demonstrated repeatedly.The embodiment of the present application can be carried out polymerization to described commodity image based on this result, by merging into a Search Results with money commodity in commercial articles searching result, shows user.No matter be identical website master, or the different main commodity that provide in website, so long as with money commodity, capital is grouped together in together, in Search Results, do once and represent, if user need to understand and have which website is main these part commodity that provide, can select described commodity to go again the inner inquiry of polymerization result.

Be appreciated that different commodity image collections can have different class indications, the class indication of all commodity images in commodity image collection can be unified.After being the unified class indication of commodity image configurations in commodity image collection, the search commercial articles request that again receives user is searched for, and obtain after traditional Search Results, can be by the class indication of identification commodity image, can fast the identical commodity image of sign be polymerized to the set of same money commodity, and need not according to its digital signature, carry out polymerization again, save system resource and dwindled polymerization time.

In addition, the embodiment of the present application can also be carried out polymerization to the commodity of new interpolation according to default time rule, be classified to accordingly with the set of money commodity, and the corresponding sign of configuration.

Wherein, described time rule can be set according to actual conditions by those skilled in the art, can be that not timing is carried out polymerization to the business object of new interpolation, such as the business object of new interpolation being carried out polymerization and the business object of new interpolation carried out to polymerization in every 6 hours at idle periods such as mornings in every 2 hours at peak hours/periods such as nights, can be periodically the business object of new interpolation to be carried out to polymerization etc., the embodiment of the present application not be limited this yet.

For the embodiment of the present application, substantially similar to the embodiment of the method for same money business object polymerization owing to obtaining the embodiment of the method for class indication, the embodiment of the present application is not described in detail in this, and relevant part is referring to the part explanation of the embodiment of the method for same money business object polymerization.

Be appreciated that, for embodiment of the method, for simple description, therefore it is all expressed as to a series of combination of actions, but those skilled in the art should know, the embodiment of the present application is not subject to the restriction of described sequence of movement, because according to the embodiment of the present application, some step can adopt other orders or carry out simultaneously.Secondly, those skilled in the art also should know, the embodiment described in instructions all belongs to preferred embodiment, and related action and module might not be that the embodiment of the present application is necessary.

With reference to Fig. 4, show a kind of same money of the embodiment of the present application with the structured flowchart of the system embodiment of money business polymerization, specifically can comprise with lower module:

Image collection module 401, for obtaining the image of business object;

Digital signature generation module 402, for generating the digital signature of described image according to the visual signature of described image;

Module 403 is set up in image signatures storehouse, for adopting described image and corresponding digital signature thereof to set up image signatures storehouse;

Module 404 is divided in grouping, for the image in image signatures storehouse being divided into a plurality of groupings according to described digital signature;

Similarity calculation module 405, for calculating the similarity of described image based on described a plurality of groupings;

Image collection forms module 406, for described image being carried out to cluster according to described similarity, forms one or more image collections;

With money business object, merge module 407, for business object corresponding to the image that belongs to same image collection merged into same money business object.

In the application's a preferred embodiment, described digital signature generation module comprises: described visual signature comprises color characteristic, and/or, Gradient Features, and/or, local feature;

Visual Feature Retrieval Process submodule, for extracting respectively the visual signature of described image;

In the application's a preferred embodiment, described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, described Visual Feature Retrieval Process submodule comprises:

And/or,

In the application's a preferred embodiment, described gradient orientation histogram calculating sub module further comprises:

In the application's a preferred embodiment, described grouping is divided module and is comprised:

In the application's a preferred embodiment, described similarity calculation module comprises:

Third phase is seemingly spent calculating sub module, for for present image, calculates the similarity of the local feature of last remaining image in main grouping that its local feature is corresponding with it and adjacent packets; Similarity is determined submodule, for the similarity using the similarity of the local feature of the main grouping corresponding with it of described present image and the last remaining image of adjacent packets as described image.

In the application's a preferred embodiment, described image collection forms module and comprises:

In the application's a preferred embodiment, described system also comprises:

In the application's a preferred embodiment, described duplicate removal module comprises:

Be appreciated that, for the system embodiment of same money business object polymerization, because it is substantially similar with the embodiment of the method for money business polymerization to same money, so description is fairly simple, relevant part is the part explanation with the embodiment of the method for money business polymerization referring to same money.

Referring to Fig. 5, show the structured flowchart of a kind of search system embodiment of the application, specifically can comprise as lower module:

Request receiving module 501, for receiving user's searching request;

Search Results acquisition module 502, for according to described request, obtains Search Results; And

Search Results merges module 503, for described Search Results being had to the business object of identical class indication, merges into same money business object;

Wherein, described class indication generates by following submodule:

Image Acquisition submodule, for obtaining the image of business object;

In the application's a preferred embodiment, described digital signature generates submodule and comprises:

And/or,

In the application's a preferred embodiment, described Search Results merges module and also comprises:

Be appreciated that for search system embodiment, because it is substantially similar to searching method embodiment, so description is fairly simple, relevant part is referring to the part explanation of searching method embodiment.

Each embodiment in this instructions all adopts the mode of going forward one by one to describe, and each embodiment stresses is the difference with other embodiment, between each embodiment identical similar part mutually referring to.

Those skilled in the art should understand, the embodiment of the embodiment of the present application can be provided as method, system or computer program.Therefore, the embodiment of the present application can adopt complete hardware implementation example, implement software example or in conjunction with the form of the embodiment of software and hardware aspect completely.And the embodiment of the present application can adopt the form that wherein includes the upper computer program of implementing of computer-usable storage medium (including but not limited to magnetic disk memory, CD-ROM, optical memory etc.) of computer usable program code one or more.

In a typical configuration, described computer equipment comprises one or more processors (CPU), input/output interface, network interface and internal memory.Internal memory may comprise the volatile memory in computer-readable medium, and the forms such as random access memory (RAM) and/or Nonvolatile memory, as ROM (read-only memory) (ROM) or flash memory (flash RAM).Internal memory is the example of computer-readable medium.Computer-readable medium comprises that permanent and impermanency, removable and non-removable media can realize information by any method or technology and store.Information can be module or other data of computer-readable instruction, data structure, program.The example of the storage medium of computing machine comprises, but be not limited to phase transition internal memory (PRAM), static RAM (SRAM), dynamic RAM (DRAM), the random access memory of other types (RAM), ROM (read-only memory) (ROM), Electrically Erasable Read Only Memory (EEPROM), fast flash memory bank or other memory techniques, read-only optical disc ROM (read-only memory) (CD-ROM), digital versatile disc (DVD) or other optical memory, magnetic magnetic tape cassette, the storage of tape magnetic rigid disk or other magnetic storage apparatus or any other non-transmission medium, can be used for the information that storage can be accessed by computing equipment.According to defining herein, computer-readable medium does not comprise the computer readable media (transitory media) of non-standing, as data-signal and the carrier wave of modulation.

The embodiment of the present application is with reference to describing according to process flow diagram and/or the block scheme of the method for the embodiment of the present application, terminal device (system) and computer program.Should understand can be in computer program instructions realization flow figure and/or block scheme each flow process and/or the flow process in square frame and process flow diagram and/or block scheme and/or the combination of square frame.Can provide these computer program instructions to the processor of multi-purpose computer, special purpose computer, Embedded Processor or other programmable data processing terminal equipment to produce a machine, the instruction of carrying out by the processor of computing machine or other programmable data processing terminal equipment is produced for realizing the device in the function of flow process of process flow diagram or a plurality of flow process and/or square frame of block scheme or a plurality of square frame appointments.

These computer program instructions also can be stored in energy vectoring computer or the computer-readable memory of other programmable data processing terminal equipment with ad hoc fashion work, the instruction that makes to be stored in this computer-readable memory produces the manufacture that comprises command device, and this command device is realized the function of appointment in flow process of process flow diagram or a plurality of flow process and/or square frame of block scheme or a plurality of square frame.

These computer program instructions also can be loaded on computing machine or other programmable data processing terminal equipment, make to carry out sequence of operations step to produce computer implemented processing on computing machine or other programmable terminal equipment, thereby the instruction of carrying out is provided for realizing the step of the function of appointment in flow process of process flow diagram or a plurality of flow process and/or square frame of block scheme or a plurality of square frame on computing machine or other programmable terminal equipment.

Although described the preferred embodiment of the embodiment of the present application, once those skilled in the art obtain the basic creative concept of cicada, can make other change and modification to these embodiment.So claims are intended to all changes and the modification that are interpreted as comprising preferred embodiment and fall into the embodiment of the present application scope.

Finally, also it should be noted that, in this article, relational terms such as the first and second grades is only used for an entity or operation to separate with another entity or operational zone, and not necessarily requires or imply and between these entities or operation, have the relation of any this reality or sequentially.And, term " comprises ", " comprising " or its any other variant are intended to contain comprising of nonexcludability, thereby the process, method, article or the terminal device that make to comprise a series of key elements not only comprise those key elements, but also comprise other key elements of clearly not listing, or be also included as the intrinsic key element of this process, method, article or terminal device.The in the situation that of more restrictions not, the key element being limited by statement " comprising ... ", and be not precluded within process, method, article or the terminal device that comprises described key element and also have other identical element.

The method of a kind of same money business object polymerization above the application being provided, the system of a kind of same money business object polymerization, a kind of searching method and a kind of search system, be described in detail, applied specific case herein the application's principle and embodiment are set forth, the explanation of above embodiment is just for helping to understand the application's method and core concept thereof; Meanwhile, for one of ordinary skill in the art, the thought according to the application, all will change in specific embodiments and applications, and in sum, this description should not be construed as the restriction to the application.

Claims

1. with a method for money business object polymerization, it is characterized in that, comprising:

Obtain the image of business object;

2. method according to claim 1, is characterized in that, the step that the described visual signature according to image generates the digital signature of described image comprises:

3. method according to claim 2, is characterized in that,

Described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, the described step of extracting respectively the visual signature of image comprises:

The color dimension that acquisition quantizes in default color space;

And/or,

4. method according to claim 3, is characterized in that, the sub-step that the gray level image after described foundation is level and smooth calculates the gradient orientation histogram of described image further comprises:

Be normalized, generate the gradient orientation histogram of present image.

5. according to the method described in claim 3 or 4, it is characterized in that, the described step that image in image signatures storehouse is divided into a plurality of groupings according to digital signature comprises:

Extract the color characteristic in the digital signature of described image;

6. method according to claim 5, is characterized in that, the described step of calculating the similarity of described image based on a plurality of groupings comprises:

7. method according to claim 6, is characterized in that, describedly according to described similarity, described image is carried out to cluster, and the step that forms one or more image collections comprises:

8. method according to claim 1, is characterized in that, described method also comprises:

9. method according to claim 8, is characterized in that, the described step that the image that simultaneously appears at the same business object in a plurality of image collections is carried out to duplicate removal processing comprises:

10. a searching method, is characterized in that, comprising:

Receive user's searching request;

According to described request, obtain Search Results; And

Wherein, the generation method of described class indication comprises:

Obtain the image of business object;

Similarity based on described a plurality of grouping computed image;

11. methods according to claim 10, is characterized in that, the step that the described visual signature according to image generates the digital signature of described image comprises:

12. methods according to claim 11, is characterized in that,

The color dimension that acquisition quantizes in default color space;

And/or,

13. methods according to claim 10, is characterized in that, the generation method of described class indication also comprises:

14. 1 kinds of systems with the polymerization of money business object, is characterized in that, comprising:

Image collection module, for obtaining the image of business object;

15. systems according to claim 14, is characterized in that, described digital signature generation module comprises:

16. systems according to claim 15, is characterized in that,

Described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, described Visual Feature Retrieval Process submodule comprises:

And/or,

17. systems according to claim 16, is characterized in that, described gradient orientation histogram calculating sub module further comprises:

18. according to the system described in claim 16 or 17, it is characterized in that, described grouping is divided module and comprised:

19. systems according to claim 18, is characterized in that, described similarity calculation module comprises:

20. systems according to claim 19, is characterized in that, described image collection forms module and comprises:

21. systems according to claim 14, is characterized in that, described system also comprises:

22. systems according to claim 21, is characterized in that, described duplicate removal module comprises:

23. 1 kinds of search systems, is characterized in that, comprising:

Request receiving module, for receiving user's searching request;

Wherein, described class indication generates by following submodule:

Image Acquisition submodule, for obtaining the image of business object;

24. systems according to claim 23, is characterized in that, described digital signature generates submodule and comprises:

25. systems according to claim 24, it is characterized in that, described color characteristic comprises the one or more main color for every image, and, the pixel number of described main color correspondence in present image, when described visual signature comprises color characteristic, described Visual Feature Retrieval Process submodule comprises:

And/or,

26. systems according to claim 23, is characterized in that, described Search Results merges module and also comprises: