Background technology
Image retrieval is meant the technology that the image in the digital picture storehouse is inquired about, and its target is to make the image that retrieves can satisfy user's intention.During retrieving images, can import the literal of rendering image content, and retrieve according to these literal; And also can import a certain image, search approximate with it image.The present invention relates to input picture, search the retrieval of approximate image.
And if the image in the image library marks, then can greatly improve the accuracy rate of retrieval.But the mark image is a job of quite wasting time and energy, if image library is too huge, then is difficult to all images are marked.
Therefore, the someone has proposed a kind of reaction type image search method, by the study feedback model, parts of images is marked, and improve constantly retrieval precision.
The present invention promptly provides a kind of reaction type image retrieval pattern, can effectively improve retrieval precision, constantly improves result for retrieval.
Summary of the invention
The present invention is directed to the drawback of prior art, a kind of feedback type approximate image retrieval system is provided, comprising: image memory device, comprising marking image storage part and not marking image storage part; Image input module to be retrieved is used to import image to be retrieved; Retrieval module, it retrieves 10 associated pictures according to the image of importing in the image input module to be retrieved to be retrieved in the mark image storage part; Feedback module feeds back to retrieval module with described 10 associated pictures, and retrieves in not marking image storage part according to these associated pictures and not mark associated picture; Display module, it is presented at and does not mark the not mark associated picture that retrieves in the image storage part; The classification labeling module, allow the user that the shown not mark associated picture branch two-stage that retrieves is marked, be about to described do not mark associated picture mark respectively with actual relevant positive sample image and with two ranks of the incoherent negative sample image of reality, and the image after will marking is transferred to the mark image storage part; Wherein, when marked reach 30 with the incoherent negative sample image of reality the time, stop the mark; Described 30 negative sample images are fed back to retrieval module by feedback module, in not marking image storage part, to retrieve the not mark image relevant with described 30 negative sample images, these images are labeled as the negative sample image automatically, and are transferred in the mark image storage part; Final retrieval sorting module, utilize retrieval module at final mark image storage part with do not mark the retrieval of carrying out image to be retrieved in the image storage part, and respectively in display module output at the mark image storage part with do not mark the final result for retrieval that retrieves in the image storage part.
Preferably, in the feedback type approximate image retrieval system of the present invention, described image memory device is the hard disk or the webserver.
Preferably, in the feedback type approximate image retrieval system of the present invention, described display module is a LCDs.
Preferably, in the feedback type approximate image retrieval system of the present invention, also comprise the selection labeling module, can select whether the final result for retrieval that shows in the display module to be marked once more for the user.
The present invention also provides a kind of reaction type image search method, and comprising following steps: image input step to be retrieved is used for to image input module input to be retrieved image to be retrieved; The primary retrieval step according to the image of importing in the image input module to be retrieved to be retrieved, retrieves 10 associated pictures by retrieval module in the mark image storage part in image memory device; A feedback step feeds back to retrieval module with described 10 associated pictures by feedback module, and does not mark associated picture according to retrieving in the not mark image storage part of these associated pictures in image memory device; Step display is presented at by display module and does not mark the not mark associated picture that retrieves in the image storage part; Annotation step, allow the user that the not mark associated picture branch two-stage that retrieves that step display shows is marked by the classification labeling module, be about to described do not mark associated picture mark respectively with actual relevant positive sample image and with two ranks of the incoherent negative sample image of reality, and the image after will marking is transferred to the mark image storage part; Wherein, when marked reach 30 with the incoherent negative sample image of reality the time, stop the mark; The secondary feedback step feeds back to retrieval module with described 30 negative sample images by feedback module; The quadratic search step retrieves the not mark image relevant with described 30 negative sample images in not marking image storage part, these images are labeled as the negative sample image automatically, and is transferred in the mark image storage part; Result for retrieval arrangement step, utilize retrieval module at final mark image storage part with do not mark the retrieval of carrying out image to be retrieved in the image storage part, and respectively in display module output at the mark image storage part with do not mark the final result for retrieval that retrieves in the image storage part.
Preferably, in the reaction type image search method of the present invention, described image memory device is the hard disk or the webserver.
Preferably, in the reaction type image search method of the present invention, described display module is a LCDs.
Preferably, in the reaction type image search method of the present invention, after described result for retrieval arrangement step, can also comprise the selection annotation step, whether it can select the final result for retrieval that shows in the display module to be marked once more by selecting labeling module for the user.
Feedback type approximate image retrieval system of the present invention and method can improve retrieval precision by twice retrieval and twice feedback, and constantly expand the mark image in the image library.
Embodiment
Below in conjunction with accompanying drawing the utility model is described in further detail, can implements according to this with reference to the instructions literal to make those skilled in the art.
As shown in Figure 1, the invention discloses a kind of feedback type approximate image retrieval system, comprising: image memory device, comprising marking image storage part and not marking image storage part.The present invention both can realize the local hard drive image retrieval, can realize the network image retrieval again.Therefore, image both can be stored in the local hard drive, also can be stored in the external server.
Also comprise image input module to be retrieved in the feedback type approximate image retrieval system, be used to import image to be retrieved.Image to be retrieved can read input by USB flash disk, also can draw by outside drawing tools and import, and also can import by the intercepting parts of images.
Also comprise retrieval module in the feedback type approximate image retrieval system, it retrieves 10 associated pictures according to the image of importing in the image input module to be retrieved to be retrieved in the mark image storage part.10 item of image that retrieve in the mark image storage part have quite high matching rate, therefore, can directly adopt usually.If but in a single day wherein there is non-matching image, the user also can delete by artificial mode.And final 10 item of image of coupling fully that form.
Also comprise feedback module in the feedback type approximate image retrieval system, described 10 associated pictures are fed back to retrieval module, and retrieve in not marking image storage part according to these associated pictures and not mark associated picture.Can enlarge range of search like this.
Also comprise display module in the feedback type approximate image retrieval system, it is presented at and does not mark the not mark associated picture that retrieves in the image storage part.And also can show result for retrieval each time, for example, be presented in the mark image storage part and retrieve 10 associated pictures, so just help the user and delete the image that do not match by manual type.
Also comprise the classification labeling module in the feedback type approximate image retrieval system, allow the user that the shown not mark associated picture branch two-stage that retrieves is marked, be about to described do not mark associated picture mark respectively with actual relevant positive sample image and with two ranks of the incoherent negative sample image of reality, and the image after will marking is transferred to the mark image storage part; Wherein, when marked reach 30 with the incoherent negative sample image of reality the time, stop the mark; Described 30 negative sample images are fed back to retrieval module by feedback module, in not marking image storage part, to retrieve the not mark image relevant with described 30 negative sample images, these images are labeled as the negative sample image automatically, and are transferred in the mark image storage part.That is to say that the not mark image that the user need not all are retrieved all marks, reach 30, remaining retrieving images is no longer marked, in order to avoid too increase user's mark beam as long as mark to the negative sample image.By feedback negative sample image, automatically some images are marked, to save manpower mark cost.
Also comprise final retrieval sorting module in the feedback type approximate image retrieval system, utilize retrieval module at final mark image storage part with do not mark the retrieval of carrying out image to be retrieved in the image storage part, and respectively in display module output at the mark image storage part with do not mark the final result for retrieval that retrieves in the image storage part.
In the described feedback type approximate image retrieval system, described image memory device is the hard disk or the webserver.Hard disk can be realized local retrieval, and the webserver can be realized network retrieval.Also can be other memory storage, USB flash disk for example, net dish etc.
In the described feedback type approximate image retrieval system, described display module is a LCDs.Also can be other screen, any display screen of the Presentation Function that human eye is visual of can realizing all can.
In the described feedback type approximate image retrieval system, also comprise the selection labeling module, can select whether the final result for retrieval that shows in the display module to be marked once more for the user.If the user finds to have obvious unmatched image in final result for retrieval, can select whether to mark.
The present invention also provides a kind of reaction type image search method, and comprising following steps: image input step to be retrieved is used for to image input module input to be retrieved image to be retrieved; The primary retrieval step according to the image of importing in the image input module to be retrieved to be retrieved, retrieves 10 associated pictures by retrieval module in the mark image storage part in image memory device; A feedback step feeds back to retrieval module with described 10 associated pictures by feedback module, and does not mark associated picture according to retrieving in the not mark image storage part of these associated pictures in image memory device; Step display is presented at by display module and does not mark the not mark associated picture that retrieves in the image storage part; Annotation step, allow the user that the not mark associated picture branch two-stage that retrieves that step display shows is marked by the classification labeling module, be about to described do not mark associated picture mark respectively with actual relevant positive sample image and with two ranks of the incoherent negative sample image of reality, and the image after will marking is transferred to the mark image storage part; Wherein, when marked reach 30 with the incoherent negative sample image of reality the time, stop the mark; The secondary feedback step feeds back to retrieval module with described 30 negative sample images by feedback module; The quadratic search step retrieves the not mark image relevant with described 30 negative sample images in not marking image storage part, these images are labeled as the negative sample image automatically, and is transferred in the mark image storage part; Result for retrieval arrangement step, utilize retrieval module at final mark image storage part with do not mark the retrieval of carrying out image to be retrieved in the image storage part, and respectively in display module output at the mark image storage part with do not mark the final result for retrieval that retrieves in the image storage part.
By twice retrieval and twice feedback, make that the result of retrieval is accurate day by day, and progressively improve the image labeling amount in the method provided by the invention, save human cost, make final result for retrieval have reference value.
In the described reaction type image search method, described image memory device is the hard disk or the webserver.
In the described reaction type image search method, described display module is a LCDs.
In the described reaction type image search method, after described result for retrieval arrangement step, can also comprise the selection annotation step, whether it can select the final result for retrieval that shows in the display module to be marked once more by selecting labeling module for the user.
Although embodiment of the present invention are open as above, but it is not restricted to listed utilization in instructions and the embodiment, it can be applied to various suitable the field of the invention fully, for those skilled in the art, can easily realize other modification, therefore under the universal that does not deviate from claim and equivalency range and limited, the legend that the utility model is not limited to specific details and illustrates here and describe.