Summary of the invention
For the above-mentioned technical problem in correlation technique, the present invention proposes a kind of recognition methods and device of geographical position.
For realizing above-mentioned technical purpose, according to an aspect of the present invention, a kind of recognition methods of geographical position is provided.
The recognition methods in this geographical position comprises:
Gather the target image containing Word message;
Target image is analyzed, extracts the Word message comprised in target image;
According to Word message and the corresponding relation between pre-configured Word message and geographical location information, determine the geographical location information corresponding with Word message.
In addition, the recognition methods in this geographical position also comprises: before gathering the target image containing Word message, detect, determine that described target image contains Word message to target image.
Wherein, Word message comprises numeral and/or alphabetical and/or Chinese character.
Further, when Word message comprises numeral and/or letter, by back-propagation algorithm, target image is detected, to determine that target image contains Word message.
Meanwhile, when Word message comprises Chinese character, by the AdaBoost algorithm based on LAB feature, target image is detected, to determine that target image contains Word message.
In addition, targeted graphical is being analyzed, when extracting the Word message comprised in targeted graphical, can analyze targeted graphical, determine whether the Word message comprised in targeted graphical comprises numeral and/or letter; And when determining that Word message comprises numeral and/or letter, then this numeral and/or letter are split, extract this numeral and/or alphabetical.
In addition, targeted graphical is being analyzed, when extracting the Word message comprised in targeted graphical, also can when determining that Word message does not comprise numeral and/or letter, secondary analysis is carried out to targeted graphical, determines whether the Word message comprised in targeted graphical comprises Chinese character; And when determining that Word message comprises Chinese character, then this Chinese character is split, extract this Chinese character.
Wherein, dividing method be can be and split by morphological image processing method.
In addition, according to Word message and the corresponding relation between pre-configured Word message and geographical location information, when determining the geographical location information corresponding with Word message, can according to the segment word information in Word message, by the corresponding relation between the first pre-configured Word message and geographical location information, pre-matching is carried out to Word message; And on prewired basis, according to other Word messages in Word message, by the corresponding relation between the second pre-configured Word message and geographical location information, Secondary Match is carried out to Word message, determine about geographical location information corresponding to Word message.
In addition, the recognition methods in this geographical position also comprises: after extracting the Word message comprised in target image, the Word message extracted is analyzed, determine the parameter information of Word message, wherein, parameter information comprise following one of at least: font information, character arrangement, text color, character background.
In addition, according to Word message and the corresponding relation between pre-configured Word message and geographical location information, when determining the geographical location information corresponding with Word message, also can on the basis of Secondary Match, according to the parameter information of Word message, by the pre-configured the 3rd corresponding relation between Word message and geographical location information, described Word message is mated again, determines the geographical location information corresponding with Word message.
According to a further aspect in the invention, a kind of recognition device of geographical position is provided.
The recognition device in this geographical position comprises:
Acquisition module, for gathering the target image containing Word message;
Extraction module, for analyzing target image, extracts the Word message comprised in target image;
Identification module, for according to Word message and the corresponding relation between pre-configured Word message and geographical location information, determines the geographical location information corresponding with Word message.
In addition, the recognition device in this geographical position also comprises: detection module, for before gathering the target image containing Word message, detects, determine that target image contains Word message to target image.
Wherein, Word message comprises numeral and/or alphabetical and/or Chinese character.
Further, when Word message comprises numeral and/or letter, detection module can be used for being detected target image by back-propagation algorithm, to determine that target image contains Word message.
Meanwhile, when Word message comprises Chinese character, detection module can be used for being detected target image by the AdaBoost algorithm based on LAB feature, to determine that target image contains Word message.
In addition, extraction module can comprise the first analysis module and the first segmentation module, and wherein, the first analysis module can be used for analyzing target image, determines whether the Word message comprised in target image comprises numeral and/or letter; First segmentation module is used in when determining that Word message comprises numeral and/or letter, splits this numeral and/or letter, to extract this numeral and/or alphabetical.
In addition, extraction module also can comprise the second analysis module and the second segmentation module, wherein, when the second analysis module is used in and determines that Word message does not comprise numeral and/or letter, secondary analysis is carried out to target image, determines whether the Word message comprised in target image comprises Chinese character; When second segmentation module is used in and determines that Word message comprises Chinese character, this Chinese character is split, to extract this Chinese character.
Wherein, dividing method be can be and split by morphological image process.
In addition, identification module can comprise the first matching module and the second matching module, and wherein, the first matching module can be used for according to the segment word information in Word message, by the corresponding relation between the first pre-configured Word message and geographical location information, pre-matching is carried out to Word message; Second matching module is used on the basis of pre-matching, according to other Word messages in Word message, by the corresponding relation between the second pre-configured Word message and geographical location information, Secondary Match is carried out to Word message, determine the geographical location information corresponding with Word message.
In addition, the recognition device in this geographical position comprises: the 3rd analysis module, for after extracting the Word message comprised in target image, the Word message extracted is analyzed, determine the parameter information of Word message, wherein, parameter information comprise following one of at least: font information, character arrangement, text color, character background.
In addition, identification module also can comprise the 3rd matching module, for the basis at Secondary Match, according to the parameter information of Word message, by the pre-configured the 3rd corresponding relation between Word message and geographical location information, Word message is mated again, determines the geographical location information corresponding with Word message.
The present invention is by analyzing the target image containing Word message obtained, determine the Word message comprised in target image, then according to Word message and the corresponding relation between pre-configured Word message and geographical location information, determine the geographical location information corresponding with Word message, thus the Word message achieved in the target image by obtaining carries out the identification in geographical position, effectively prevent when the band of position residing for user can not receiving satellite signal time and can not carry out Geographic mapping situation occur, and then improve the experience property of user.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, the every other embodiment that those of ordinary skill in the art obtain, all belongs to the scope of protection of the invention.
According to embodiments of the invention, provide a kind of recognition methods of geographical position.
As shown in Figure 1, comprise according to the recognition methods in the geographical position of the embodiment of the present invention:
Step S101, gathers the target image containing Word message;
Step S103, analyzes target image, extracts the Word message comprised in target image;
Step S105, according to Word message and the corresponding relation between pre-configured Word message and geographical location information, determines the geographical location information corresponding with Word message.
In addition, the recognition methods in this geographical position also comprises: before gathering the target image containing Word message, detect, determine that described target image contains Word message to target image.
Wherein, Word message comprises numeral and/or alphabetical and/or Chinese character.
Further, when Word message comprises numeral and/or letter, by back-propagation algorithm, target image is detected, to determine that target image contains Word message.
Meanwhile, when Word message comprises Chinese character, by the AdaBoost algorithm based on LAB feature, target image is detected, to determine that target image contains Word message.
In addition, targeted graphical is being analyzed, when extracting the Word message comprised in targeted graphical, can analyze targeted graphical, determine whether the Word message comprised in targeted graphical comprises numeral and/or letter; And when determining that Word message comprises numeral and/or letter, then this numeral and/or letter are split, extract this numeral and/or alphabetical.
In addition, targeted graphical is being analyzed, when extracting the Word message comprised in targeted graphical, also can when determining that Word message does not comprise numeral and/or letter, secondary analysis is carried out to targeted graphical, determines whether the Word message comprised in targeted graphical comprises Chinese character; And when determining that Word message comprises Chinese character, then this Chinese character is split, extract this Chinese character.
Wherein, dividing method be can be and split by morphological image processing method.
In addition, according to Word message and the corresponding relation between pre-configured Word message and geographical location information, when determining the geographical location information corresponding with Word message, can according to the segment word information in Word message, by the corresponding relation between the first pre-configured Word message and geographical location information, pre-matching is carried out to Word message; And on prewired basis, according to other Word messages in Word message, by the corresponding relation between the second pre-configured Word message and geographical location information, Secondary Match is carried out to Word message, determine about geographical location information corresponding to Word message.
In addition, the recognition methods in this geographical position also comprises: after extracting the Word message comprised in target image, the Word message extracted is analyzed, determine the parameter information of Word message, wherein, parameter information comprise following one of at least: font information, character arrangement, text color, character background.
In addition, according to Word message and the corresponding relation between pre-configured Word message and geographical location information, when determining the geographical location information corresponding with Word message, also can on the basis of Secondary Match, according to the parameter information of Word message, by the pre-configured the 3rd corresponding relation between Word message and geographical location information, described Word message is mated again, determines the geographical location information corresponding with Word message.In such scheme, for the detection of Word message, it detects based on grader in fact, therefore, when practical application, before the detection carrying out Word message, needs the image extracting fixed size at random as sample, with training classifier.But although the sample extracted at random is identical in size, but, still there is certain difference between the sample of identical or different object, and this species diversity main manifestations is on the unity and coherence in writing of sample, such as, when sample position and angle different, also can there is larger difference in the unity and coherence in writing of each sample.
Therefore, in order to reduce difference between sample and sample module inter-object distance, accelerate the convergence rate of training algorithm, when practical application, preliminary treatment can be carried out to the sample graphics gathered, such as, manual mark sample characteristics position and/or by carrying out geometric correction (comprising rotation, dimension correction) to sample image.
When practical application, for being detected target image by back-propagation algorithm in such scheme, to determine the numeral that target image contains and/or letter, can first by the sample image containing numeral and letter, split according to fixed size (such as, 20 × 20cm), and undertaken indicating (character A by the title of set form, available A-001-XXX or A-002-XXX, waits form to indicate); Adopt gray feature to carry out histogrammic equalization processing to sample image simultaneously.And then train with BP (Back-propagation, back-propagation algorithm), obtain the grader of each character or letter.And after obtaining corresponding grader, then can to the target image photographed, with different proportional zooms repeatedly (such as, 4 times), and to the picture obtained at every turn according to fixed size (such as, 20 × 20cm) search for, to detect numeral in target image and/or letter, and after retrieving corresponding numeral and/or word, the picture region detected is marked, when numeral near the same area and/or letter exceed certain threshold value, then can merge existence numeral and/or alphabetical region, obtain the position at its place.
Accordingly, when practical application, AdaBoost algorithm based on LAB feature is detected target image, to determine that target image contains Chinese character, LAB first can be adopted to be characterized as and to describe the substantially special of object, with RealAdaBoost algorithm construction strong classifier, this strong classifier of structure is then utilized to detect target image.Wherein, when constructing strong classifier, grader cascade structure also can be used to provide detection speed and accuracy rate, DiscreteAdaBoost algorithm also can be utilized from all possible feature to select 8 structural feature LAB features, to increase the spatial description ability of feature simultaneously.
In addition, when practical application, Chinese character is detected, word size just can detect in the scope of 12x12cm size, therefore, adopt based on the AdaBoost algorithm of LAB feature, Chinese character is detected time, training requirement level is 7 layers just can obtain desirable Detection results, and then can be embedded in chip easily, reach the object of fast detecting.
In addition, when practical application, the morphological image processing method that utilizes in such scheme is split, its main method first finds and letter and/or digital and/or that Chinese character is corresponding region, then to this region found, Canny filtering is adopted, with the textural characteristics of outstanding relevant range target, and the area image obtained is adopted to the algorithm of accumulation, accentuated edges feature.And technique scheme for a better understanding of the present invention, below in conjunction with the principle process schematic diagram of the recognition methods in geographical position, technique scheme of the present invention is described in detail.
Fig. 2 is the principle process schematic diagram of the recognition methods in geographical position, as can be seen from Figure 2, when carrying out geographical position and identifying, first the image sequence containing landmark information (i.e. Word message) is gathered, then text detection is carried out to the image sequence gathered, when there is Word message, then Word message is processed, and according to the Word message after process, one-level terrestrial reference storehouse (the first namely pre-configured Word message and the corresponding relation between geographical location information) is utilized once to mate Word message, and after Word message is once mated, recycling secondary terrestrial reference storehouse, Secondary Match is carried out on the basis of once mating, find the geographical location information corresponding with Word message, and show this geographical location information.
Wherein, for the process of Word message, comprise in fact the text detection in natural background and the Text segmentation in natural background.
For the text detection in natural background, when detecting, word may be exist with the formation that full Chinese character or full word are female or digital, also be likely mix with letter with Chinese character or Chinese character mixes with numeral or the alphabetical and digital form mixed exists, and for this kind of situation, when detecting, corresponding detection method is adopted to carry out detecting successively, such as, when word be with Chinese character and letter form exist time, BP algorithm first can be utilized to detect letter, then detect Chinese character with the AdaBoost algorithm based on LAB feature.
In addition, for the detection of Chinese character, when practical application, Chinese character also may exist with the formation of different fonts, arrangement mode, color and background, and such as, the font of the imitative regular script of standard that may adopt had exists, also the formation of the Song typeface or other font that may adopt had exists, and for this kind of situation, when carrying out the training of actual classification device, corresponding training is carried out to grader.
And for the Text segmentation in natural background, first can adopt the method for statistical learning, obtain the font information of word, determine the position of word, and then obtain the segmentation result of word by the method for morphological image.Wherein, if the difference on corresponding word indicating location (such as, some words are erect-type, some words are horizontal tablet formulas), then need the positional information of other words determined except current character, to determine that word exists in position in what manner, and then carry out the segmentation of word.
Wherein, for the coupling of Word message, be in fact by individual character, double word or three words, carry out the coupling of keyword in different forms.Such as, " Shi Ning mansion ", the last character " summer " is obtained by Text segmentation, utilize arest neighbors method and " one-level terrestrial reference storehouse ", determine " mansion ", then enter in " secondary terrestrial reference storehouse ", " secondary terrestrial reference " corresponding to " mansion ", utilize other Word messages that segmentation obtains, adopt statistical learning method, finally obtain " Shi Ning mansion ".Wherein, for " one-level terrestrial reference storehouse " and " secondary terrestrial reference storehouse ", the geographical location information corresponding to each keyword is wherein comprised.
In such scheme, only relate to and by " one-level terrestrial reference storehouse " and " secondary terrestrial reference storehouse ", geographical location information has been positioned, and in fact, in order to improve the accuracy of location, also can position geographical location information according to " three grades of terrestrial reference storehouse (not shown) ".
Concrete, according to the parameter information of Word message (such as, the parameter information that font information, character arrangement, text color, character background etc. are relevant) carry out corresponding relation between configured letters information and geographical location information (namely the 3rd corresponding relation) between Word message and geographical location information, then, the parameter information confirmed according to analyzing Word message, carries out refinement to the position that Secondary Match draws.Such as, " Peking University ", blue background is the west gate of Peking University, and white background is the south gate of Peking University, the east gate of vertical setting of types Ze Shi Peking University.
As can be seen here, by such scheme of the present invention, the identification in geographical position can be carried out by the Word message in the target image of acquisition, meet the demand that user carries out running fix.
In addition, according to embodiments of the invention, a kind of recognition device of geographical position is additionally provided.
As shown in Figure 3, comprise according to the recognition device in the geographical position of the embodiment of the present invention:
Acquisition module 31, for gathering the target image containing Word message;
Extraction module 32, for analyzing target image, extracts the Word message comprised in target image;
Identification module 33, for according to Word message and the corresponding relation between pre-configured Word message and geographical location information, determines the geographical location information corresponding with Word message.
In addition, the recognition device in this geographical position also comprises: detection module (not shown), for before gathering the target image containing Word message, detects, determine that target image contains Word message to target image.
Wherein, Word message comprises numeral and/or alphabetical and/or Chinese character.
Further, when Word message comprises numeral and/or letter, detection module can be used for being detected target image by back-propagation algorithm, to determine that target image contains Word message.
Meanwhile, when Word message comprises Chinese character, detection module can be used for being detected target image by the AdaBoost algorithm based on LAB feature, to determine that target image contains Word message.
In addition, extraction module 32 can comprise the first analysis module (not shown) and the first segmentation module (not shown), wherein, the first analysis module can be used for analyzing target image, determines whether the Word message comprised in target image comprises numeral and/or letter; First segmentation module is used in when determining that Word message comprises numeral and/or letter, splits this numeral and/or letter, to extract this numeral and/or alphabetical.
In addition, extraction module 32 also can comprise the second analysis module (not shown) and the second segmentation module (not shown), wherein, when second analysis module is used in and determines that Word message does not comprise numeral and/or letter, secondary analysis is carried out to target image, determines whether the Word message comprised in target image comprises Chinese character; When second segmentation module is used in and determines that Word message comprises Chinese character, this Chinese character is split, to extract this Chinese character.
Wherein, dividing method be can be and split by morphological image process.
In addition, identification module 33 can comprise the first matching module (not shown) and the second matching module (not shown), wherein, first matching module can be used for according to the segment word information in Word message, by the corresponding relation between the first pre-configured Word message and geographical location information, pre-matching is carried out to Word message; Second matching module is used on the basis of pre-matching, according to other Word messages in Word message, by the corresponding relation between the second pre-configured Word message and geographical location information, Secondary Match is carried out to Word message, determine the geographical location information corresponding with Word message.
In addition, the recognition device in this geographical position comprises: the 3rd analysis module (not shown), for after extracting the Word message comprised in target image, the Word message extracted is analyzed, determine the parameter information of Word message, wherein, parameter information comprise following one of at least: font information, character arrangement, text color, character background.
In addition, identification module also can comprise the 3rd matching module (not shown), for the basis at Secondary Match, according to the parameter information of Word message, by the pre-configured the 3rd corresponding relation between Word message and geographical location information, Word message is mated again, determines the geographical location information corresponding with Word message.
Recognition device due to the above-mentioned geographical position in the present invention is corresponding with the RM in geographical position, therefore, just sets forth no longer one by one in this principle for the recognition device in geographical position.
In sum, by means of technique scheme of the present invention, by analyzing the target image containing Word message obtained, determine the Word message comprised in target image, then according to Word message and the corresponding relation between pre-configured Word message and geographical location information, determine the geographical location information corresponding with Word message, thus the Word message achieved in the target image by obtaining carries out the identification in geographical position, effectively prevent when the band of position residing for user can not receiving satellite signal time and can not carry out Geographic mapping situation occur, and then improve the experience property of user.
In addition, by means of technique scheme of the present invention, by carrying out multistage coupling to Word message, can realize accurately and obtaining corresponding geographical location information fast, and then effectively raising recognition rate and the efficiency in geographical position.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.