A kind of image searching method based on visual signature
(1) technical field:
The present invention relates to identification and the search technique of picture, particularly relate to the selection of picture feature variable and the analysis to picture feature spectral line and extracting method and the application in identification and the search of picture thereof.
(2) background technology:
Based on visual signature or content-based picture searching technical research from first business-like content-based image with dynamically scene searching system---the QBIC of IBM Corporation has had the history of more than ten years till now.
Now similar techniques main on our times is done to an introduction:
1.QBIC (Query By Image Content) image indexing system is image and the dynamic scene searching system of the IBM Corporation's exploitation and composition nineties, is first content-based business-like image indexing system.QBIC system provides multiple inquiry mode, comprise: utilize standard model figure (self provides system) retrieval, user draws sketch or scanning input picture is retrieved, and selects color or structure query mode, and user inputs the object retrieval moving in motion video fragment and prospect.In the time of user's input picture, sketch or video fragment, QBIC carries out the features such as color, texture, shape to the query image of input and carries out analysis and drawing out, and the inquiry mode of then selecting according to user carries out respectively different processing.The color characteristic colored number percent, the color position distribution etc. that in QBIC, use; The textural characteristics using is that the one representing according to the texture of Tamura proposition is improved, and combines the characteristic of roughness, contrast and directivity; The shape facility using has area, circularity, degree of eccentricity, main shaft deflection and one group of algebraically square invariant.QBIC or a few have been considered one of system of indexing of high dimensional features.QBIC, except the retrieval of content-based characteristic above, is also aided with text query means.
2.Virage is the CBIR engine of being developed by Virage company. the same with QBIC system, it also supports the image retrieval based on visual signatures such as color, color layout, texture and structures.
VIR (the Visual Information Retrieval) image engine of VIRAGE company provides four kinds of visual attributes retrievals (color, composition, texture and shape).Every kind of attribute is endowed 0 to 10 weights.It is the most simple and clear retrieving by color characteristics, and tone, color and the degree of saturation of this software to the base image of selecting analyzed, and then in image library, searches and the immediate image of these color attributes.Composition (composition) characteristic refers to the degree of approximation in relevant colors region.User can set one or more attribute weights and optimize retrieval.Reaching optimum balance degree needs repetition test, but retrieving is quickish.In result display matrix, can select to check 3,6,9,12,15 or 18 sketches.By the adjustment to four attribute weights, demonstrate different result for retrieval.Sketch is according to similarity descending sort.Click sketch title by obtaining some detailed descriptions of this image, comprise the ratio of similitude that Virage calculates.
3.RetrievalWare is a kind of CBIR instrument of being developed by Excalibur Science and Technology Ltd..In earlier version, can see this system focus on use neural network algorithm to realize image retrieval.In newer version, r provides the retrieval based on 6 kinds of image attributes, is respectively color, shape, texture, color structure, brightness structure and aspect ratio.Color attribute is that color to image and shared ratio thereof are measured, but does not comprise structure to color or the mensuration of position, and this is by color structure property control; Shape attribute refers to the profile of objects in images or the relative orientation of lines, flexibility and contrast; Texture properties refers to smoothness or the roughness of image, the character of surface of a width figure; Brightness attribute refers to the brightness of the pixel combination of composing images.
4.Photobook be the multi-media Laboratory of Massachusetts Institute Technology develop for image querying and the interactive tool browsed.It is made up of three subsystems, is responsible for respectively extracting shape, texture, facial characteristics.Therefore, user can carry out respectively based on shape, based on texture and the image retrieval based on facial characteristics in these three subsystems.
5.VisualSEEK is the gopher based on visual signature, and WebSEEK is a kind of text towards WWW or image search engine.These two searching systems are all developed by Columbia University.Their principal feature is the visual signature that has adopted spatial relationship between image-region and extracted from compression domain.The visual signature that system adopts is to utilize color set and the textural characteristics based on wavelet transformation.VisualSEEK supports the inquiry based on visual signature and the inquiry based on spatial relationship simultaneously.WebSEEK comprises three main modular: image/video acquisition module, subject classification and index module, search, browse and retrieval module.
Without exception, these based on visual signature or content-based picture searching technology in, texture and shape are two kinds of different attributes.It is higher that complicated algorithm and structure require the structure of knowledge of the user to using these technology.Algorithm complexity, data processing amount is bigger than normal, and the many features of manual intervention are also apparent to the cost pressure of large-scale commercial applications operation.
(3) summary of the invention:
Technical matters to be solved by this invention is: in the identification and search of picture, select suitable picture feature variable and picture feature characteristics of variables spectral line is analyzed, adopt the Eigenvalue Extraction Method that reduces deal with data amount, reduce the popular universal difficulty using.
Because client process data volume is little, easy operating, can be widely used in the field such as internet photographic search engine, mobile terminal picture search.
Owing to can arbitrarily determining according to user intention effective coverage and the content of search, can also be used for the picture searching field of subscriber's local computing machine again.
For solving above technical matters, the present invention is disclosing following technical scheme.
(4) embodiment:
A realization for image searching method based on visual signature, comprising:
By characteristic variable quantification assignment such as the form and aspect of picture, saturation degree, brightness, gray scale and planimetric positions.
Transfer the picture file in picture library to standard thumbnail according to setting physical dimension.
Obtain form and aspect, saturation degree, brightness, gray scale and the plane positional number value of the each pixel of standard thumbnail, form a property data base of standard thumbnail single features variable.
Standard thumbnail, according to the different accuracy of identification analyses of setting, is formed to the transition boundary line of single features variable.And formed the characteristic spectral line of the single features variable of each picture file standard thumbnail by whole transition boundary lines.All the single features characteristics of variables spectral line of picture file standard thumbnail forms the quadratic character database of this characteristic variable.
On single features variable transition boundary line, the tangential direction of each pixel position forms three property data bases of this characteristic variable.
The full property data base of property data base, quadratic character database and three property data base formation picture library picture standard thumbnail.Full property data base is associated with the URL (Uniform Resoure Locator) of picture file in picture library.
When search, on the picture as sample file, the one or more continuous or discontinuous curve that passes through " search target " of obtaining using computer entry device is as " search condition ".Common factor to " search condition " with transition boundary line, the numerical value such as the tangential direction according to its form and aspect, saturation degree, brightness, gray scale, planimetric position and transition boundary line on this pixel (point bunch) are compared with full property data base, complete single features variable or many characteristic variables combinatorial search.
Search result be returned as with " search condition " degree of agreement meet predefined picture with and URL.
The following vocabulary relating in the technical program refers to:
" picture library ": the picture that the picture storage device of local computer or search engine can grab in network.
" standard thumbnail ": the fixed measure thumbnail definitely according to search accuracy and file size balance.
" transition boundary line ": according to the accuracy of identification of single features variable, the mid point that numerical value change pixel line the occurs smooth curve forming that is linked in sequence.Because the pixel expression in current display technique causes transition boundary line not dropped on any pixel, when actual treatment, two curves that form with the pixel of both sides, transition boundary line calculate respectively.