CN102436578A - Formation method for dog face characteristic detector as well as dog face detection method and device - Google Patents

Abstract

The invention discloses a formation method for a dog face characteristic detector as well as a dog face detection method and device. The formation method for a dog face characteristic detector comprises the following steps of: extracting a first preset quantity of dog face characteristic images; extracting a second present quantity of images which do not contain the dog face characteristic, wherein the second preset quantity is bigger than the first preset quantity; taking the dog face characteristic images as a positive sample of an Adaboost method; and taking the images which do not contain the dog face characteristic as the negative sample of the Adaboost method to obtain the dog face characteristic detector. The dog face characteristic detector formed by the technical scheme has good performance and can quickly and accurately detect the dog face characteristic. With the dog face detection method disclosed by the technical scheme of the invention, the position on which the dog face is positioned can be accurately detected, and the dog face detection method has high accuracy and good instantaneity when being used for detecting the dog face.

Description

The formation method of dog face property detector, dog face detecting method and device

Technical field

The present invention relates to area of pattern recognition, relate in particular to a kind of formation method, dog face detecting method and device of dog face property detector.

Background technology

Target detection also makes target extract, and is the image segmentation of how much of a kind of based targets and statistical nature, with target cut apart and identification unites two into one.In complex scene, need handle in real time a plurality of targets usually, therefore, the particular importance that just seems is extracted and discerned to target automatically.Along with developing rapidly of computer vision technique; It is widely used in object detection field, for now, has had the multiple method that detects to specific objective in the prior art; As: the detection of people's face, pedestrian detection, vehicle detection etc.; And from the development of target detection, the research aspect the detection of people's face, pedestrian detection, vehicle detection is also extensive, and in the research of dog face context of detection seldom.

As everyone knows, dog is human Achates, at present; In China, the U.S., Japan and other countries; The people who supports dog increases gradually, and therefore, the image processing techniques relevant with dog has extensive market prospects; Especially dog face recognition technology has played crucial effects at aspects such as dog registration, dog authentication, the making of dog photograph album and public safety monitoring.

It is a very important link in the identification of dog face that the dog face detects, and it also belongs to the detection to specific objective, and the purpose that the dog face detects detects the dog face exactly from image background, in other words conj.or perhaps the subwindow of the dog face subwindow with non-dog face is separated.For people's face detected, the dog face detects wanted the many of difficulty, this mainly be because: the dog face is various in vary such as shape, size, hair color, attitude, expressions; Dog ears and face become mildewed and block its face easily, especially the eyes part; When taking dog, owing to all be in the state of walking about as the one of which, need to capture, the position of dog face in image is always uncertain in other words very greatly so cause the change in location of dog face in image, and people's face then is usually located at the center of image; The image difference that factor such as change of background, illumination variation causes; The dog strains class is more, and the dog sample image is difficult for gathering, and above-mentioned reason has caused existing object detection method directly effectively not detect the dog face, and therefore, how can correctly detect the dog face becomes one of present problem demanding prompt solution.

The technology of relevant target detection can be CN101799875A referring to publication number, and denomination of invention is a kind of one Chinese patent application of object detection method, but for the problems referred to above, it does not relate to.

Summary of the invention

The problem that the present invention solves is how to detect the dog face accurately.

For addressing the above problem, the present invention provides a kind of formation method of dog face property detector, comprises the steps:

Extract the dog face characteristic image of first predetermined quantity;

Extract the image of not comprising of second predetermined quantity of said dog face characteristic; Said second predetermined quantity is greater than said first predetermined quantity;

With the positive sample of said dog face characteristic image as the Adaboost method, the said image that does not comprise said dog face characteristic obtains said dog face property detector as the negative sample of Adaboost method.

Optional, extract dog face characteristic image and comprise:

From training image, extract the image that comprises dog face characteristic;

The said image that comprises dog face characteristic is carried out normalization, obtain dog face characteristic image.

Optional, the said image that comprises dog face characteristic that from training image, extracts comprises:

Demarcate the upper left corner and the position/upper right corner, the lower right corner and the position, the lower left corner of dog face characteristic in the said training image;

The center that obtains dog face characteristic in the said training image based on the upper left corner of said dog face characteristic and position/upper right corner, the lower right corner and position, the lower left corner;

Center with said dog face characteristic is a symcenter; The upper left corner of moving said dog face characteristic and position/upper right corner, the lower right corner and position, the lower left corner are wide identical until the upper left corner of said dog face characteristic and the lower right corner/upper right corner and lower left corner distance in the horizontal direction and said dog face characteristic, high identical in the distance of vertical direction and said dog face characteristic.

Optional, the said image that comprises dog face characteristic that from training image, extracts comprises:

Foundation comprises the rectangular window of said dog face characteristic;

Adjust the size and the position of said rectangular window, circumscribed until the edge of adjusted rectangular window and said dog face characteristic.

Optional, saidly the said image that comprises dog face characteristic is carried out normalization comprise:

Normalization size and the said size of images that comprises dog face characteristic based on predetermined are obtained the convergent-divergent multiple;

Based on said convergent-divergent multiple, the said image that comprises dog face characteristic is carried out interpolation.

Optional, when the said image that comprises dog face characteristic is coloured image, the said gray level image that comprises the image of dog face characteristic is carried out normalization.

Optional, the formation method of said dog face property detector also comprises: upset about the image that comprises dog face characteristic after the normalization is carried out.

Optional, the said upset about the image that comprises dog face characteristic after the normalization is carried out obtains through following mode:

$\left\{\begin{array}{c}{P}_1(i,j)=P(i,n_c-j+1)\\ 1\&le;i\&le;{\mathrm{<figure-callout\; id="1"\; label="n\; r"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">n</figure-callout>}}_r\\ 1\&le;j\&le;n_c\end{array}\right.$

Wherein, P (i, n _c-j+1) be the capable n of i in the image that comprises dog face characteristic after the normalization _cThe pixel value of-j+1 row, P ₁(i, j) be about the pixel value of the capable j of i row in the image after the upset, n _r* n _cIt is the size of images that comprises dog face characteristic after the normalization.

Optional, said with the positive sample of said dog face characteristic image as the Adaboost method, the said image that does not comprise said dog face characteristic obtains said dog face property detector as the negative sample of Adaboost method and comprises:

Training parameter in the said Adaboost method is set, and said training parameter comprises: positive sample number, negative sample number, just/negative sample normalization size, progression, the positive sample percent of pass of each grade minimum and the maximum false alarm rate of each grade;

Obtain said dog face property detector based on said training parameter, said positive sample, negative sample.

Optional, said dog face is characterized as the dog nostril, and said dog face property detector is a dog nostril detecting device, and the training parameter in the said Adaboost method is:

Said positive sample number is more than or equal to 500;

Said negative sample number is 1.5～5 times of said positive sample number;

Be of a size of 20 * 20～28 * 28 pixels after just said/negative sample normalization;

Said progression is 15～25 grades;

The positive sample percent of pass of said each grade minimum is 0.995～0.998;

The maximum false alarm rate of said each grade is 0.46～0.50.

Optional, said dog face is characterized as dog eyes, and said dog face property detector is the dog eye detector, and the training parameter in the said Adaboost method is:

Said positive sample number is more than or equal to 400;

Said negative sample number is 2～4 times of said positive sample number;

Just said/negative sample normalization is of a size of 10 * 15～20 * 30 pixels;

Said progression is 15～25 grades;

The positive sample percent of pass of said each grade minimum is 0.995～0.998;

The maximum false alarm rate of said each grade is 0.46～0.50.

For addressing the above problem, the present invention also provides a kind of dog face detecting method, comprising:

Treat detected image and carry out pre-service;

Adopt the dog face property detector of above-mentioned acquisition that the dog face characteristic in the pretreated image to be detected is detected;

Testing result through said dog face property detector is demarcated said dog face.

Optional, the said detected image of treating is carried out pre-service and is comprised:

Based on the segmentation zoom factor said image to be detected is carried out convergent-divergent, said segmentation zoom factor is associated with the size of dog face property detector;

To the image rotation predetermined angular to be detected behind the convergent-divergent.

Optional, the said detected image of treating is carried out pre-service and is also comprised:

After said image to be detected is carried out convergent-divergent, before the rotation of the image to be detected behind convergent-divergent predetermined angular, upset about the image to be detected behind the said convergent-divergent carried out.

Optional, the dog face property detector of the above-mentioned acquisition of said employing detects the dog face characteristic in the pretreated image to be detected and comprises:

Size according to the current search window is cut apart at least one search subwindow of acquisition to pretreated image to be detected;

Detect said search subwindow through said dog face property detector, until detecting the search subwindow that has said dog face characteristic;

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than pretreated size of images to be detected;

There is the position of the search subwindow of said dog face characteristic in output.

Optional, said dog face is characterized as the dog nostril, and said dog face property detector is a dog nostril detecting device, and said testing result through said dog face property detector is demarcated said dog face and comprised through following formula and demarcate said dog face:

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ y_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

Wherein, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

Optional, said dog face is characterized as dog eyes, and said dog face property detector is the dog eye detector, and said testing result through said dog face property detector is demarcated said dog face and comprised through following formula and demarcate said dog face:

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=\min (y_{a1},y_{b1})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

Wherein, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

For addressing the above problem, the present invention also provides a kind of dog face pick-up unit, comprising:

The dog face property detector of above-mentioned acquisition;

Pretreatment unit is suitable for treating detected image and carries out pre-service;

Detecting unit is suitable for adopting said dog face property detector that the dog face characteristic in the pretreated image to be detected is detected;

Demarcate the unit, be suitable for said dog face being demarcated based on the testing result of said detecting unit.

For addressing the above problem, the present invention also provides a kind of dog face detecting method, comprising:

Treat detected image and carry out pre-service;

Adopt the dog nostril detecting device of above-mentioned acquisition that the dog nostril in the pretreated image to be detected is detected;

The dog eyes that adopt the dog eye detector of above-mentioned acquisition to treat in the detected image detect;

Testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face.

Optional, the said detected image of treating is carried out pre-service and is comprised:

Based on the segmentation zoom factor said image to be detected is carried out convergent-divergent, said segmentation zoom factor is associated with the size of dog nostril detecting device;

To the image rotation predetermined angular to be detected behind the convergent-divergent.

Optional, the said detected image of treating is carried out pre-service and is also comprised:

After said image to be detected is carried out convergent-divergent, before the rotation of the image to be detected behind convergent-divergent predetermined angular, upset about the image to be detected behind the convergent-divergent carried out.

Optional, the dog nostril detecting device of the above-mentioned acquisition of said employing detects the dog nostril in the pretreated image to be detected and comprises:

Size according to the current search window is cut apart at least one search subwindow of acquisition to pretreated image to be detected;

Detect said search subwindow through said dog nostril detecting device, until detecting the search subwindow that has said dog nostril;

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than pretreated size of images to be detected;

There is the position of the search subwindow in said dog nostril in output.

Optional, the dog eye detector of the above-mentioned acquisition of said employing is treated dog eyes in the detected image and is detected and comprise:

Treat detected image according to the size of current search window and cut apart at least one search subwindow of acquisition;

Detect said search subwindow through said dog eye detector, until detecting the search subwindow that has said dog eyes;

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than size of images to be detected;

There is the position of the search subwindow of said dog eyes in output.

Optional, said testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face and is comprised:

When detecting dog nostril and two dog eyes, and the dog nostril do not overlap with two dog eyes, then through following formula said dog face demarcated:

$\left\{\begin{array}{c}{x}_a=0.5\&times;(x_{a2}+x_{a1})\\ y_a=0.5\&times;(y_{a2}+y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_b=0.5\&times;(x_{b2}+x_{b1})\\ y_b=0.5\&times;(y_{b2}+y_{b1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_c=0.5\&times;(x_{c2}+x_{c1})\\ y_c=0.5\&times;(y_{c2}+y_{c1})\end{array}\right.$

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_a-0.5\&times;(x_b-x_a)\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=\min (y_a,y_b)-0.25\&times;(y_c-\min (y_a,y_b))\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_b+0.5\&times;(x_b-x_a)\\ y_{f2}=y_c+0.5\&times;(y_c-\min (y_a,y_b))\end{array}\right.$

Wherein, (x _a, y _a) be the center of the left eye eyeball of calibrated dog, (x _b, y _b) be the center of the right eye eyeball of calibrated dog, (x _c, y _c) be the center in calibrated dog nostril, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

Optional, said testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face and is comprised:

Do not overlap with dog eyes when detecting dog nostril and dog eyes and said dog nostril; Perhaps only detect the dog nostril, then said dog face demarcated through following formula:

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

Wherein, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

Optional, said testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face and is comprised:

When detecting dog nostril and two dog eyes, and said dog nostril overlaps with dog eyes, then through following formula said dog face demarcated:

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=\min (y_{a1},y_{b1})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

Wherein, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

For addressing the above problem, the present invention also provides a kind of dog face pick-up unit, comprising:

The dog nostril detecting device of above-mentioned acquisition;

The dog eye detector of above-mentioned acquisition;

Pretreatment unit is suitable for treating detected image and carries out pre-service;

Dog nostril detecting unit is suitable for adopting said dog nostril detecting device that the dog nostril in the pretreated image to be detected is detected;

Dog eye detection unit, the dog eyes that are suitable for adopting said dog eye detector to treat in the detected image detect;

Demarcate the unit, be suitable for said dog face being demarcated based on the testing result of said dog nostril detecting unit and said dog eye detection unit.

Compared with prior art, technical scheme of the present invention has the following advantages:

Through with dog face characteristic image as the positive sample of Adaboost method, do not comprise the negative sample of the image of said dog face characteristic as the Adaboost method; And then train the performance of dog face property detector of acquisition good through the Adaboost method, can detect the characteristic of dog face accurately and rapidly.

Through adopting above-mentioned dog face property detector that the dog face in the pretreated image to be detected is detected, can detect the position at dog face place exactly, and the accuracy when said dog face detected is high, real-time is good.

Further; Said pre-service comprises that treating detected image carries out rotating predetermined angular again behind the segmentation convergent-divergent; Make when adopting said dog face property detector that the dog face is detected; Can detect the relatively poor image to be detected of picture quality; And when treating detected image and carrying out adopting said dog face property detector that the dog face is detected again after the above-mentioned pre-service, can also accelerate having the detection speed of high-resolution image to be detected, and then realize the dog face in the different images to be detected is carried out accurate detection.

Further; Through dog nostril detecting device the dog nostril in the pretreated image to be detected is detected, the dog eyes of treating in the detected image through the dog eye detector detect, and it is main having realized detecting with the dog nostril; The dog eye detection detects the dog face for assisting; Because be main with the dog nostril in the testing process, the dog eyes are auxilliary, so can remove the detection false-alarm in the dog nostril that possibly hit on the dog eyes; And then can realize detection more accurately and rapidly to the dog face; And with the testing result of dog nostril detecting device is that the testing result of master, dog eye detector is auxilliary the dog face to be detected, and real-time accuracy good, that detect is high, and false alarm rate is very low.

Description of drawings

Fig. 1 is the process flow diagram of formation method of the dog face property detector of embodiment of the present invention;

Fig. 2 is the process flow diagram of formation method of the dog nostril detecting device of the embodiment of the invention one;

Fig. 3 is the synoptic diagram of the extraction dog nostril image of the embodiment of the invention one;

Fig. 4 is the process flow diagram of formation method of the dog eye detector of the embodiment of the invention two;

Fig. 5 is the process flow diagram of the dog face detecting method of the embodiment of the invention three;

Fig. 6 treats detected image to carry out pretreated process flow diagram in the embodiment of the invention three;

Fig. 7 is the location diagram of dog nostril and dog eyes in the dog face model;

Fig. 8 is the actual example figure after adopting the dog face detecting method of the embodiment of the invention three to treat detected image to detect;

Fig. 9 is the structural representation of the dog face pick-up unit of the embodiment of the invention three;

Figure 10 is the process flow diagram of the dog face detecting method of the embodiment of the invention four;

Figure 11 is the structural representation of the dog face pick-up unit of the embodiment of the invention four.

Embodiment

As what describe in the background technology, almost blank in the prior art for dog face context of detection, and the dog face detects for people's face detects relatively more difficult.

The inventor is through discovering, with respect to other pets as: for cat, the rabbit etc., the differentiation property in the nostril of dog is relatively good; The kind of dog how no matter this is, most dog all has circular nostril, and promptly the dog nostril is more stable on shape and structure; And for general dog, the dog nostril is bigger than dog eyes usually, and the dog nostril is not subject to blocking of its face's hair color; The influence that changed by angle is little, and promptly the robustness of angle is good.And for the dog eyes, under the situation of not blocking, it is also relatively stable on shape and structure.

Therefore; The inventor proposes to form dog face property detector through the mode of off-line training; Said dog face characteristic can be dog nostril, dog eyes, dog face, be used for the different characteristic of dog face is carried out relevant detection, among the present invention particularly; Form dog nostril detecting device the dog nostril is detected, form the dog eye detector dog eyes are detected.

Be based on the dog face property detector that obtains in the off-line training process, and then in the process of online detection, utilize said dog face property detector that the dog face is detected, demarcate the position at dog face place.Particularly, through the dog nostril in the online detection of the dog nostril detecting device image to be detected, and then dog face position is demarcated through the position in detected dog nostril; Perhaps through the dog eyes in the online detection of the dog eye detector image to be detected, and then through the position of detected dog eyes dog face position is demarcated; Perhaps not only detected the dog nostril in the image to be detected but also detected the dog eyes in the image to be detected through the dog eye detector through dog nostril detecting device; And then be main with the position in the detected dog of detecting device nostril, dog nostril; The position of the detected dog eyes of dog eye detector is auxilliary; Treat the position of the dog face in the detected image and demarcate, to realize accurate detection the dog face.

See also Fig. 1, Fig. 1 is the process flow diagram of formation method of the dog face property detector of embodiment of the present invention, and as shown in Figure 1, the formation method of said dog face property detector comprises:

Step S1: the dog face characteristic image that extracts first predetermined quantity.

Step S2: the image that extracts not comprising of second predetermined quantity of said dog face characteristic; Said second predetermined quantity is greater than said first predetermined quantity.

Step S3: with the positive sample of said dog face characteristic image as the Adaboost method, the said image that does not comprise said dog face characteristic obtains said dog face property detector as the negative sample of Adaboost method.

For the formation method of dog face property detector that embodiment of the present invention is described better, the formation method of the dog face property detector of embodiment of the present invention is carried out detailed explanation below in conjunction with the formation method of dog nostril detecting device and the formation method of dog eye detector.

Embodiment one

See also Fig. 2, Fig. 2 is the process flow diagram of formation method of the dog nostril detecting device of the embodiment of the invention one, and as shown in Figure 2, the formation method of said dog nostril detecting device comprises:

Step S1a: the dog nostril image that extracts first predetermined quantity.

Step S2a: extract the image that does not comprise said dog nostril of second predetermined quantity, said second predetermined quantity is greater than said first predetermined quantity.

Step S3a: with the positive sample of said dog nostril image as the Adaboost method, the said image that does not comprise said dog nostril obtains said dog nostril detecting device as the negative sample of Adaboost method.

Execution in step S1a, in order to extract the dog nostril image of first predetermined quantity, (image that comprises dog that will gather usually is called training and uses the dog image to need elder generation to gather the image that comprises dog of some; The image in the dog nostril of extracting is called training with dog nostril image); In the present embodiment, particularly, can pass through image capture device; As camera, video camera or on the net the search packet picture that contains dog gather training and use the dog image; In order to improve the performance of dog nostril detecting device, the training of collection with the number of dog image should try one's best many, the training of collection both can be that coloured image also can be a black white image with the dog image.

In the present embodiment; The training that collects can be identical with said first predetermined quantity with the quantity of dog image; Also can be slightly less than said first predetermined quantity (this situation be applicable to a width of cloth training use the dog image) with the training that comprises the image of at least two dogs in the dog image; Preferably, the training that collects is identical with said first predetermined quantity with the quantity of dog image, the said first predetermined quantity N ₁More than or equal to 500 width of cloth; And in order to extract clear and complete dog nostril image; Each width of cloth for collecting is trained for the dog image; The front that should as far as possible comprise the dog face of a dog, the perhaps approximate front of the dog face of a dog can be clearly visible fully with the nostril of guaranteeing dog.

In the present embodiment, in order to detect the nostril of most of dog, the training that collects is with the wide W in dog nostril in the dog image _nShould be more than or equal to 22 pixels, the high H in dog nostril _nShould be more than or equal to 22 pixels, promptly the size in dog nostril should satisfy:

$\left\{\begin{array}{c}{W}_n\&GreaterEqual;22\\ H_n\&GreaterEqual;22\end{array}\right.$

Further; Consider if training is too big with the size in the dog nostril in the dog image; The nostril that causes the dog nostril detecting device of follow-up acquisition to treat doggie in the detected image probably detect less than; And if training undersized with the dog nostril in the dog image, then possibly cause the dog nostril detecting device of follow-up acquisition to treat dog nostril in the detected image when detecting, differentiate the nostril of unclear dog and cause omission or flase drop.Therefore, in the present embodiment, the training that collects is preferably with the size in dog nostril in the dog image:

$\left\{\begin{array}{c}22\&le;W_n\&le;26\\ 22\&le;H_n\&le;26\end{array}\right.$

The training that utilization collects is extracted dog nostril image with the dog image, comprises particularly:

With extracting the image that comprises the dog nostril the dog image, the said image that comprises the dog nostril is carried out normalization from training, obtain dog nostril image.

In the present embodiment, can carry out through following dual mode with extracting the image that comprises the dog nostril the dog image, below explain one by one from training.

First kind:

Demarcate the upper left corner and lower right corner position/upper right corner and the lower left corner position of said training with dog nostril in the dog image.

Obtain the center of said training based on the upper left corner in said dog nostril and position/upper right corner, the lower right corner and position, the lower left corner with dog nostril in the dog image.

Center with said dog nostril is a symcenter; The upper left corner of moving said dog nostril and position/upper right corner, the lower right corner and position, the lower left corner are wide identical until the upper left corner in said dog nostril and the lower right corner/upper right corner and lower left corner distance in the horizontal direction and said dog nostril, high identical in the distance of vertical direction and said dog nostril.

In the present embodiment, describe as example to demarcate the upper left corner and the position, the lower right corner that said training uses dog nostril in the dog image.See also Fig. 3, Fig. 3 is the synoptic diagram of the extraction dog nostril image of the embodiment of the invention one, and is as shown in Figure 3, and the position, the upper left corner of demarcating the dog nostril is initial position (x _N1, y _N1), position, the lower right corner is final position (x _N2, y _N2), in order to obtain the image in complete dog nostril, need guarantee with the line between said initial position and the said final position to be that cornerwise rectangle should comprise the dog nostril fully.After the upper left corner in dog nostril and position, the lower right corner were demarcated, the center in dog nostril can obtain through following formula:

$\left\{\begin{array}{c}{x}_n=\frac{x_{n2}+x_{\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">n</figure-callout>}1}}{2},\\ y_n=\frac{y_{n2}+{\mathrm{<figure-callout\; id="1"\; label="y\; n"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_n}{2},\end{array}\right.$

Then, be symcenter with the center in said dog nostril, move the position, the upper left corner and the position, the lower right corner in said dog nostril.

Particularly, the wide W between setting initial position (position, the upper left corner in dog nostril) and the final position (position, the lower right corner in dog nostril) ₁=x _N2～x _N1, the high H between initial position and the final position ₁=y _N1-y _N2, the horizontal coordinate x of maintenance final position _N2Constant, Mobile Termination position vertically keeps the ordinate y of initial position _N1Constant, along continuous straight runs moves initial position, is the position of symcenter adjustment initial position and final position then with the center, until W ₁Wide identical with the dog nostril, H ₁High identical with the dog nostril promptly can extract dog nostril image.

Above-mentioned is that the upper left corner in dog nostril, the lower right corner that final position is the dog nostril are that example is illustrated to demarcate initial position only; Can also demarcate the upper right corner that initial position is the dog nostril in other embodiments, dog nostril image is extracted in the lower left corner that final position is the dog nostril; Concrete leaching process no longer launches to detail with the initial position to be that the process of dog nostril image is extracted in the upper left corner in dog nostril, the lower right corner that final position is the dog nostril similar here.

Second kind:

Foundation comprises the rectangular window in said dog nostril.

Adjust the size and the position of said rectangular window, circumscribed until the edge in adjusted rectangular window and said dog nostril.

Particularly, training with the dog image in foundation comprise the rectangular window in dog nostril, the length breadth ratio of the rectangular window of preferentially being set up is identical with the ratio of width to height in said dog nostril, the size of dwindling rectangular window is circumscribed until the edge in the four edges of rectangular window and dog nostril.

Usually, it is circular that the dog nostril is approximately, therefore; The ratio approximately equal of the height in the wide and dog nostril in dog nostril; Usually the scope of the ratio of width to height in dog nostril is 0.9～1.1, in actual detected, for the ease of gathering training with dog nostril image; Usually when the image to the dog nostril extracts, be 1 to extract the image in dog nostril with the ratio of width to height in dog nostril.

In addition, in the present embodiment, if the training that collects uses the dog image to be coloured image; The dog nostril image that then extracts also is a coloured image; Need be converted into gray level image this moment, particularly, the coloured image in dog nostril converted into the gray level image in dog nostril through following formula:

P＝0.299×R+0.587×G+0.114×B

Wherein, P is a gray values of pixel points arbitrarily in the gray level image in dog nostril after the conversion, and R, G, B are respectively the component values of this pixel corresponding red, green and blue primary in the coloured image in dog nostril.

Through above-mentioned steps from the training with after having extracted the image that comprises the dog nostril the dog image; Also need carry out normalization to the image that comprises the dog nostril that extracts handles; The dog nostril size of images that comprises through extracting is obtained corresponding convergent-divergent multiple with predetermined normalization size; Come the image that comprises the dog nostril that extracts is carried out interpolation based on said convergent-divergent multiple and image difference method, and then realize normalization said dog nostril image.

In the present embodiment, the dog nostril size of images after the normalization can be between 20 * 20～28 * 28 pixels, below are 24 * 24 pixels with the dog nostril size of images after the normalization, carry out briefly bright to the normalization process of said dog nostril image.

The convergent-divergent multiple RN that obtains dog nostril image according to the physical size of extracting that comprises dog nostril image and the normalized size of ultimate demand ₁, said convergent-divergent multiple RN ₁Concrete through following formula acquisition:

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="RN"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">RN</figure-callout>}}_1=\frac{W_{\mathrm{Nn}}}{W_{\mathrm{In}}}\\ W_{\mathrm{Nn}}=24\end{array}\right.$ Or $\left\{\begin{array}{c}{\mathrm{RN}}_{}\end{array}\right.$ $\left\{\begin{array}{c}{}_1=\frac{H_{\mathrm{Nn}}}{H_{\mathrm{In}}}\\ H_{\mathrm{Nn}}=24\end{array}\right.$

Wherein, W _InWide for the dog nostril image that extracts, H _InBe the height of the dog nostril image that extracts, W _NnWide for the dog nostril image after the normalization, H _NnHeight for the dog nostril image after the normalization.

Convergent-divergent multiple RN based on the dog nostril image that obtains ₁With corresponding image interpolation method the image in dog nostril is carried out normalization; Said image interpolation method can adopt bilinear interpolation (bilinearinterpolation), arest neighbors interpolation (nearest neighbor interpolation), bicubic interpolation (bicubicinterpolation) etc.; In the present embodiment, preferably adopt bilinear interpolation method or arest neighbors interpolation method that the dog nostril image that extracts is carried out normalization.The convergent-divergent multiple that adopts image interpolation method and image is normalized to the known technology of this area to image, so, no longer launch specifically detailed description here.

After the image to said dog nostril carries out normalization; In order to obtain the more sample number; Performance with the dog nostril detecting device that improves follow-up acquisition; Upset obtains new training with dog nostril image about can carrying out the dog nostril image after the normalization that obtains, so for the above-mentioned extraction first predetermined quantity N ₁The image in dog nostril, through about upset, can obtain 2N ₁Dog nostril image.

Particularly, through following formula dog nostril image is carried out about the upset:

$\left\{\begin{array}{c}{P}_1(i,j)=P(i,n_c-j+1)\\ 1\&le;i\&le;{\mathrm{<figure-callout\; id="1"\; label="n\; r"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">n</figure-callout>}}_r\\ 1\&le;j{\&le;n}_c\end{array}\right.$

Wherein, P (i, n _c-j+1) be the capable n of i in the dog nostril image after the normalization _cThe pixel value of-j+1 row, P ₁(i, j) be about the pixel value of the capable j of i row in the dog nostril image after the upset, n _r* n _cIt is the dog nostril size of images after the normalization.Owing in the present embodiment, be that 24 * 24 pixels describe with the dog nostril size of images after the normalization, so n _r=24, n _c=24.

So far, through said method from the training that collects with having extracted the image that comprises the dog nostril the dog image, obtained the positive sample that following adopted Adaboost training method training obtains dog nostril detecting device.

Execution in step S2a: extract the image that does not comprise said dog nostril of second predetermined quantity, in this step, both can never comprise the image that extracts non-dog nostril in the image of dog, also can from the image that comprises dog, extract non-dog nostril image.The image that the said image that does not comprise dog perhaps comprises dog can pass through camera, camera acquisition, perhaps the method through internet searching obtains.

In order to improve the performance of dog nostril detecting device; Especially improve the differentiation performance of the dog nostril detecting device pair target similar with said dog nostril; In the present embodiment, preferably, the training that never comprises dog is with the image that extracts non-dog nostril in the image; And the said training that does not comprise dog is with comprising various common backgrounds and other more targets, for example: cat, rabbit, deer, tiger, lion, pedestrian and icon or the like as far as possible in the image.And the quantity of the image that does not comprise the dog nostril that collects can be identical with second predetermined quantity; Also can be less than said second predetermined quantity; For latter event, then be through in each width of cloth image, extracting several images that do not comprise the dog nostril to obtain the image that does not comprise the dog nostril of second predetermined quantity.

The dog nostril size of images of extracting in the present embodiment that does not comprise after dog nostril size of images and the normalization is identical.Extracting the image that does not comprise the dog nostril is the common practise of this area, so, no longer launch to give unnecessary details here.

In addition, need to prove, if the training of extracting uses the image that does not comprise the dog nostril also to be coloured image; Then still need convert thereof into and be gray level image; Particularly, the formula that adopts in the time of can converting the gray level image in dog nostril into referring to above-mentioned coloured image with the dog nostril repeats no more here.

In the present embodiment, the said second predetermined quantity N ₂Greater than the said first predetermined quantity N ₁, preferably, the said second predetermined quantity N ₂At 1.5 N ₁～5 N ₁Between.

So far, obtained the negative sample of following adopted Adaboost training method training acquisition dog nostril detecting device through step S2a.

Execution in step S3a: be utilized in the training that obtains among the step S1a with dog nostril image as the positive sample in the Adaboost method; The training that in step S2a, obtains, is trained to obtain dog nostril detecting device as the negative sample in the Adaboost method with the image that does not comprise the dog nostril.Adopting Adaboost method training detecting device is the common practise of this area, no longer launches to give unnecessary details here.In the present embodiment, the training parameter in the Adaboost method comprises: positive sample number, negative sample number, just/negative sample normalization size, progression, the positive sample percent of pass of each grade minimum and the maximum false alarm rate of each grade.

Said positive sample number is the number of the training of said extracted with dog nostril image, can know about dog nostril image carried out can be 2N so locate the number of positive sample so that the number of dog nostril image doubles after the upset by above-mentioned ₁, said negative sample number, the training that is above-mentioned extraction is with the number that does not comprise dog nostril image; Said progression is the included multistage number of Adaboost detecting device, and it all has the relevant detection performance index for each level.The positive sample percent of pass of each level is meant for each level; Correct detected positive sample number accounts for the number percent of total positive sample number; For instance; The dog nostril image that extracts is 100 width of cloth, and it is dog nostril image that the dog nostril detecting device that training at the corresponding levels obtains detects 60 width of cloth, and then the positive sample percent of pass of this level is 60%; Each grade false alarm rate is meant for each level; Error-detecting to the negative sample number account for the number percent of total positive sample number; For instance; Having extracted the image that 100 width of cloth comprise the dog nostril, is the images that do not comprise the dog nostril but the dog nostril detecting device that training at the corresponding levels obtains detects 20 width of cloth, and then the false alarm rate of this level is 20%.For the more excellent dog nostril detecting device of obtained performance, when training parameter was provided with, the percent of pass of the minimum positive sample of each grade should be high as much as possible, and the maximum false alarm rate of each grade should be low as much as possible.

In the present embodiment; In order to make the better performances of the dog nostril detecting device that finally obtains; Need carry out corresponding setting to the training parameter in the Adaboost method; Particularly, positive sample number is more than or equal to 500, i.e. training is greater than with the number of dog nostril image or equals 500 width of cloth (training of actual acquisition can be more than or equal to 250 width of cloth with the dog image; Extracting the image that comprises the dog nostril more than or equal to the training of 250 width of cloth in the dog image more than or equal to 250 width of cloth, about carrying out again upset final obtain training with the number of dog nostril image more than or equal to 500 width of cloth); The negative sample number is 1.5～5 times of said positive sample number, i.e. training uses the number that does not comprise dog nostril image for training 1.5～5 times with dog nostril picture number; Just/negative sample normalization is of a size of 20 * 20～28 * 28 pixels; It is 15～25 grades that progression is used in training; The positive sample percent of pass of each grade minimum is 0.995～0.998; The maximum false alarm rate of each grade is 0.46～0.50.In the present embodiment, preferably, the concrete setting of correlation parameter sees also table 1 when adopting Adaboost method training dog nostril detecting device.

Table 1

Positive sample number	1564	Progression	19
				The negative sample number	3000	The positive sample percent of pass of each grade minimum	0.998
The normalization size	24×24	The maximum false alarm rate of each grade	0.48

So far, obtained dog nostril detecting device through above-mentioned step S1a, S2a, S3a.

Embodiment two

See also Fig. 4, Fig. 4 is the process flow diagram of formation method of the dog eye detector of the embodiment of the invention two, and as shown in Figure 4, the formation method of said dog eye detector comprises:

Step S1b: a dog eye image that extracts first predetermined quantity.

Step S2b: extract the image that does not comprise said dog eyes of second predetermined quantity, said second predetermined quantity is greater than said first predetermined quantity.

Step S3b: with the positive sample of said dog eye image as the Adaboost method, the said image that does not comprise said dog eyes obtains said dog eye detector as the negative sample of Adaboost method.

In the present embodiment; Among the step S1b, extract a dog eye image of first predetermined quantity, similar with the process of the dog nostril image that extracts first predetermined quantity among the embodiment one; Different is with the dog image from training; Extraction comprises the image of dog eyes, so locate the image that extracts dog eyes is no longer launched concrete detailed description, only simply explains.

In the present embodiment, for the extraction of a dog eye image, the said first predetermined quantity N ₁(training of actual acquisition can be more than or equal to 100 width of cloth with the dog image more than or equal to 400 width of cloth; The training of gathering with the dog image in left eye and/or the eye image of dog of extraction dog; Upset about the left-eye image of the dog of extracting carried out, upset then can be extracted the image more than or equal to dog eyes of 400 width of cloth about the eye image of the dog of extracting carried out).

Consider can detect the dog eyes as far as possible accurately when following adopted dog eye detector detects the dog eyes, thus the training that collects with in the dog image, the wide W of arbitrary dog eyes _eShould satisfy more than or equal to 15 pixels the high H of arbitrary dog eyes _eShould satisfy more than or equal to 10 pixels, promptly the size of arbitrary dog eyes should satisfy:

$\left\{\begin{array}{c}{W}_e\&GreaterEqual;15\\ H_e\&GreaterEqual;10\end{array}\right.$

Further; Consider if training is too big with the size of the dog eyes in the dog image; Cause probably eye detection that the dog eye detector of follow-up acquisition treats doggie in the detected image less than; And if training undersized with the dog eyes in the dog image, in the time of then possibly causing the dog eye detector of follow-up acquisition to treat dog eyes in the detected image detecting, differentiate the eyes of unclear dog and cause omission or flase drop.Therefore, in the present embodiment, the training that collects is preferably with the size of arbitrary dog eyes in the dog image:

$\left\{\begin{array}{c}13\&le;W_e\&le;17\\ 8{\&le;H}_e\&le;12\end{array}\right.$

The training that utilization collects is extracted a dog eye image with the dog image, comprises particularly:

With extracting the image that comprises dog eyes the dog image, the said image that comprises dog eyes is carried out normalization from training, obtain a dog eye image.In the present embodiment, similar from training from training with the process of extracting the image that comprises the dog nostril the dog image with extracting the dog image among a dog eye image and the embodiment one, also can adopt like a kind of described dual mode of embodiment and carry out.

During extraction; As long as will collect to such an extent that training uses the upper left corner/upper right corner demarcation of dog eyes that meet size in the dog image to be final position as initial position, the lower right corner/lower left corner demarcation; Confirm the center of dog eyes then according to calibrated initial position and final position; And then be that symcenter moves said initial position and final position with said center; Wide identical until said initial position and final position distance in the horizontal direction and said dog eyes, high identical at the distance of vertical direction and said dog eyes.

When adopting the second way from train, to extract the dog eyes with the dog image; Particularly; In training, set up the rectangular window that comprises dog eyes with the dog image; The length breadth ratio of the rectangular window of preferably being set up is identical with the ratio of width to height of dog eyes, and the size of dwindling rectangular window is circumscribed until the edge of the four edges of rectangular window and dog eyes.

Usually, the scope of the ratio of width to height of dog eyes is 1.3～1.7, in actual detected, use the dog eye image for the ease of gathering training, usually to the image extraction of dog eyes the time, is 1.5 to extract the image of dog eyes with the ratio of width to height of dog eyes.

Need to prove; In the present embodiment; If the training of extracting uses the dog eye image to be coloured image, then also need be converted into gray level image, dog eyes coloured image converts the formula that gray level image can convert gray level image referring to the dog nostril coloured image among the embodiment one into into to carry out.

Behind the image that has extracted the dog eyes; Also need carry out normalization to the image of dog eyes handles; In the present embodiment; The size of the dog eyes after the normalization can be between 10 * 15～20 * 30 pixels, and the image of said dog eyes is carried out normalization, also are to obtain corresponding convergent-divergent multiple through the size of extracting that comprises the dog eye image with predetermined normalization size; Based on said convergent-divergent multiple and image interpolation method the dog eye image that extracts is carried out interpolation, and then realize normalization said dog eye image.

After the image to said dog eyes carries out normalization, still can obtain more training and use the dog eye image, and then improve the performance of the dog eye detector of follow-up acquisition through upset about the dog eye image after the normalization is carried out.So for the above-mentioned extraction first predetermined quantity N ₁The image of dog eyes, through about upset, can obtain 2N ₁The dog eye image.

Upset can be carried out referring to the formula of upset about among the embodiment one dog nostril image being carried out about training after the normalization carried out with the dog eye image, if in the present embodiment, the dog eye image after the normalization is of a size of 10 * 15, n in the then above-mentioned formula _r=10, n _c=15.

In the present embodiment; The image that does not comprise said dog nostril that extracts second predetermined quantity among the step S2b among the image that does not comprise said dog eyes of extraction second predetermined quantity and the embodiment one is similar; Size of extracting that does not comprise the dog eye image and the dog eye image after the normalization measure-alike, and the said second predetermined quantity N ₂Greater than the said first predetermined quantity N ₁, preferably, the said second predetermined quantity N ₂At 2N ₁～4N ₁Between.

In the present embodiment; Among the step S3b with the training that obtains among the step S1b with the dog eye image as the positive sample in the Adaboost method; The training that obtains among the step S2b uses the image that does not comprise said dog eyes as the negative sample in the Adaboost method; Adopt and adopt Adaboost method training dog nostril detecting device similar among Adaboost method training dog eye detector and the embodiment one; Being provided with of different is in training process training parameter is different, in the present embodiment, in order to make the better performances of the dog eye detector finally obtained; Being provided with as follows of relevant training parameter when adopting the Adaboost method that the dog eye detector is trained: positive sample number is more than or equal to 400, promptly trains number with the dog eye image to be greater than or equals 400 width of cloth; The negative sample number is 2～4 times of said positive sample number, i.e. training uses the quantity that does not comprise the dog eye image for training 2～4 times with dog eye image quantity; Just/negative sample normalization is of a size of 10 * 15～20 * 30 pixels; It is 15～25 grades that progression is used in training; The positive sample percent of pass of each grade minimum is 0.995～0.998; The maximum false alarm rate of each grade is 0.46～0.50.In the present embodiment, preferably, the concrete setting of correlation parameter sees also table 2 when adopting Adaboost method training dog eye detector.

Table 2

Positive sample number	600	Progression	20
				The negative sample number	1500	The positive sample percent of pass of each grade minimum	0.996
The normalization size	10×15	The maximum false alarm rate of each grade	0.48

So far, obtained the dog eye detector through above-mentioned step S1b, S2b, S3b.

After said method acquisition dog nostril detecting device and dog eye detector; Then can pass through these two kinds of dog face property detectors; Treating the position at dog nostril and the dog eyes place in the detected image detects; And then demarcate the position of dog face through the position at detected dog nostril and dog eyes place, promptly detect the dog face.In order to specify the dog face detecting method of the embodiment of the invention, below detect the dog face and come the dog face detecting method of the embodiment of the invention is explained accordingly to detect dog face, dog eye detector through dog nostril detecting device respectively through the mode of dog nostril detecting device and dog eye detector joint-detection dog face.

Embodiment three

In the present embodiment, be to be main with the detected dog of detecting device nostril, dog nostril, the detected dog eyes of dog eye detector detect the dog face for assisting.

See also Fig. 5, Fig. 5 is the process flow diagram of the dog face detecting method of the embodiment of the invention three, and is as shown in Figure 5, and said dog face detecting method comprises:

Step S4: treat detected image and carry out pre-service.

Step S5: adopt dog nostril detecting device that the dog nostril in the pretreated image to be detected is detected.

Step S6: the dog eyes that adopt the dog eye detector to treat in the detected image detect.

Step S7: the testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face.

Execution in step S4; In order can to detect to the dog face in the different images to be detected; As dog nostril less in the second-rate image to be detected is detected; Also in order to accelerate high-resolution image processing velocity to be detected, at first treat detected image and carry out corresponding pre-service simultaneously.See also Fig. 6, Fig. 6 treats detected image to carry out pretreated process flow diagram in the embodiment of the invention three, and as shown in Figure 6, said pre-service comprises:

Step S4a: based on the segmentation zoom factor said image to be detected is carried out convergent-divergent, said segmentation zoom factor is associated with the size of dog nostril detecting device.

Step S4b: upset about the image to be detected behind the convergent-divergent carried out.

Step S4c: to the image rotation predetermined angular to be detected behind the convergent-divergent.

Execution in step S4a, particularly, first structural segmentation convergent-divergent function, in the present embodiment, segmentation convergent-divergent function is specially:

$f\left(x\right)=\left\{\begin{array}{cc}{s}_1& x\&Element;(0,a_1)\\ \&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;& \\ s_n& x\&Element;[a_{n-1},a_n)\\ \&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;& \\ s_{N-1}& x\&Element;[a_{N-2},a_{N-1})\\ s_N=\frac{1280}{x}& x\&Element;[a_{N-1},+\&infin;)\end{array}\right.$

Wherein, N is total interval number, s _nBe n interval corresponding convergent-divergent multiple, 1≤n≤N; The value of N is relevant with size of images to be detected, and the N value is the positive integer between 3～15 in the present embodiment; X is a size bigger in the size of images to be detected, and particularly, if image to be detected is horizontal figure, then x is the wide of horizontal figure, if image to be detected is perpendicular figure, then x is the height of perpendicular figure.In the present embodiment, the interval division of convergent-divergent is in particular: a ₁Value between 50～100, a _N-1Value is between 800～3000, and other a _nValue can be by increasing progressively picked at random or by formula: 2≤n≤N-2 carries out; The value of convergent-divergent multiple is specially: s ₁Value between 5～15, s _N-1Value is 1, s _NValue does

Other s _nValue can be by successively decreasing picked at random or by formula: 2≤n≤N-2.

For image to be detected, if W is the wide of image to be detected, H is high, establishes m and be the parameter when choosing corresponding convergent-divergent multiple s, then obtains m through following formula:

M=α * W+ β * H, α >=0 wherein, β >=0, and alpha+beta=1

Then with the independent variable of m as the piecewise function f of above-mentioned structure, promptly the x in piecewise function this moment is m, belongs to the different multiple s that detected image should be carried out convergent-divergent that select to treat when interval at m.

After having obtained the corresponding convergent-divergent multiple on the different intervals through above-mentioned segmentation convergent-divergent function; Treat detected image based on said convergent-divergent multiple and combining image interpolation method and carry out corresponding convergent-divergent, so that can detect the dog nostril in the image to be detected accurately through said dog nostril detecting device.Said image interpolation method can adopt like bilinear interpolation method, arest neighbors interpolation method, bicubic interpolation method etc., preferably adopts bilinear interpolation method or arest neighbors interpolation method in the present embodiment.

In the present embodiment; The pairing convergent-divergent multiple of piecewise function of structure is relevant with the size of the dog nostril detecting device that training obtains; Concrete, said convergent-divergent multiple is decided by the size of dog nostril size in the image to be detected of statistics gained on the size of the dog nostril detecting device of training gained and each convergent-divergent interval.The inventor has finally confirmed the interval convergent-divergent multiple that reaches corresponding to said segmentation convergent-divergent interval of more excellent segmentation convergent-divergent through several images to be detected that collect are experimentized, and in the present embodiment, the segmentation convergent-divergent interval of said segmentation is 12 intervals, particularly, is followed successively by (0; 100), [100,150), [150,200), [200; 250), [250,300), [300,350), [350; 400), [400,450), [450,500), [500; 800), [800,1280) and [1280 ,+∞),, adopt above-mentioned segmentation convergent-divergent function corresponding to corresponding interval; Both can obtain each interval convergent-divergent multiple of going up correspondence is 6,5.5,5,4.5,4,3.5 successively; 3,2.5,2,1.5,1,

In the present embodiment; Accuracy when detecting in order to improve the dog face of treating in the detected image further; Preferably, to upset about carrying out through pretreated image to be detected, said about in the formation method of upset and the formation method of dog nostril detecting device or dog eye detector about upset similar; So, no longer launch concrete the detailed description here.

Because the dog face in the image to be detected is not necessarily its front, so the dog nostril also is not necessarily the front, the accuracy when adopting dog nostril detecting device to detect the dog nostril in order to improve is to the image rotation predetermined angle pretreated to be detected after the upset about process.And because the dog nostril be circular, so the robustness of its angle variation is good, to the right or image pretreated to be detected to the left for the dog nostril; Can through to through pre-service with about after the upset image to be detected clockwise or be rotated counterclockwise predetermined angle α so that the dog nostril more near positive, particularly, if the dog nostril to the right; α then turns clockwise; If the dog nostril to the left, then is rotated counterclockwise α, the value of α is spent between 10 degree 5 in the present embodiment.

After image to be detected carried out above-mentioned pre-service; Promptly to said image to be detected carried out convergent-divergent, about the upset, the rotation after; All execution in step S5 adopts the dog nostril detecting device that obtains among the embodiment one that the dog nostril in the pretreated image to be detected is detected.Particularly,

Size according to the current search window is cut apart at least one search subwindow of acquisition to pretreated image to be detected.

Detect said search subwindow through said dog nostril detecting device, until detecting the search subwindow that has said dog nostril.

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than pretreated size of images to be detected.

There is the position of the search subwindow in said dog nostril in output.

In the present embodiment, said preset multiple is relevant with pretreated size of images to be detected, in general; Said preset multiple is between 1.2～1.5, and the original dimension of current search window and dog nostril detecting device measure-alike is of a size of 24 * 24 pixels with above-mentioned dog nostril detecting device; Said preset multiple is 1.2 for example, and the original dimension that the current search window then is set is 24 * 24 pixels, according to the original dimension of current search window pretreated image to be detected is cut apart; Obtain a plurality of search subwindows; Adopt dog nostril detecting device that it is detected to each search subwindow, as detect the dog nostril, the position of then storing this search subwindow; And output exists the position of the search subwindow in said dog nostril, detection of end.If do not detect the dog nostril; (according to preset multiple should be 28.8 * 28.8 then the size of current search window to be updated to 29 * 29 pixels; Detect for ease; So the size of the current search window after upgrading should be the size of current search window and the product of preset multiple are rounded), based on the size of the current search window after upgrading, continue pretreated image to be detected is cut apart; Search dog nostril, the size of the current search window after final updated is greater than pretreated size of images to be detected.

After adopting dog nostril detecting device that the dog nostril in the pretreated image to be detected is detected, the dog eyes that adopt the dog eye detector to treat in the detected image detect.Particularly, treat detected image according to the size of current search window and cut apart at least one search subwindow of acquisition.

Detect said search subwindow through said dog eye detector, until detecting the search subwindow that has said dog eyes.

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than size of images to be detected.

There is the position of the search subwindow of said dog eyes in output.

When adopting the dog eye detector to detect the dog eyes in the image to be detected, said preset multiple is relevant with size of images to be detected, in general; Said preset multiple is between 1.2～1.5, and the original dimension of current search window and dog eye detector measure-alike is of a size of 10 * 15 pixels with the dog eye detector; Preset multiple is 1.2 for example, and the original dimension that the current search window then is set is 10 * 15 pixels, treats detected image according to the original dimension of current search window to cut apart; Obtain a plurality of search subwindows; Adopt the dog eye detector that it is detected to each search subwindow, as detect the dog eyes, the position of then storing this search subwindow; And output exists the position of the search subwindow of said dog eyes, detection of end.If do not detect the dog eyes; Then the size with the current search window is updated to 12 * 18 pixels; Size based on the current search window after upgrading; Continue to treat detected image and cut apart, search dog eyes, the size of the current search window after final updated is greater than size of images to be detected.In addition; If in renewal process, the size of the current search window after the renewal is not an integer, when above-mentioned current search window 12 * 18 pixels are upgraded again; Then be 14.4 * 21.6; Likewise, the size of the current search window after the renewal should be the size of current search window and the product of preset multiple are rounded, so the size of the search window after the renewal of current search window 12 * 18 pixels should be 15 * 22.

Dog nostril and the dog eyes that promptly can treat in the detected image through above-mentioned dog nostril detecting device and dog eye detector detect, and then can demarcate the dog face through the testing result of dog nostril and dog eyes.

Before the dog face is demarcated, earlier the position relation of dog nostril and dog eyes in the dog face model is explained accordingly.See also Fig. 7, Fig. 7 is the location diagram of dog nostril and dog eyes in the dog face model; As shown in Figure 7; A and B represent the center of right eye eyeball of left eye eyeball and the dog of dog respectively among the figure; C represents the center in dog nostril, and after the position, the upper left corner of a dog left side/right eye eyeball and position, the lower right corner were confirmed, the center of a dog left side/right eye eyeball then can be confirmed; After the position, the upper left corner in dog nostril and position, the lower right corner were confirmed, the center in dog nostril had also just been confirmed.Center through dog left eye eyeball and right eye eyeball or/and the center in dog nostril promptly can demarcate the dog face.

In this enforcement, after employing dog nostril detecting device and dog eye detector are carried out joint-detection, just can demarcate the dog face through the testing result of dog nostril detecting device and dog eye detector.Particularly, in testing process following testing result possibly appear:

(1) detect dog nostril and two dog eyes simultaneously, and dog nostril and two dog eyes do not overlap, then the testing result of the testing result in dog nostril and dog eyes is correctly; And then demarcate through the dog face that following formula is treated in the detected image:

$\left\{\begin{array}{c}{x}_a=0.5\&times;(x_{a2}+x_{a1})\\ y_a=0.5\&times;(y_{a2}+y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_b=0.5\&times;(x_{b2}+x_{b1})\\ y_b=0.5\&times;(y_{b2}+y_{b1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_c=0.5\&times;(x_{c2}+x_{c1})\\ y_c=0.5\&times;(y_{c2}+y_{c1})\end{array}\right.$

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_a-0.5\&times;(x_b-x_a)\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=\min (y_a,y_b)-0.25\&times;(y_c-\min (y_a,y_b))\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_b+0.5\&times;(x_b-x_a)\\ y_{f2}=y_c+0.5\&times;(y_c-\min (y_a,y_b))\end{array}\right.$

Wherein, (x _a, y _a) be the center of the left eye eyeball of calibrated dog, (x _b, y _b) be the center of the right eye eyeball of calibrated dog, (x _c, y _c) be the center in calibrated dog nostril, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

(2) detect dog nostril and dog eyes simultaneously, and said dog nostril and dog eyes do not overlap, then the testing result in dog nostril is correct, and the testing result of dog eyes is a false-alarm; Perhaps only detect the dog nostril, then said dog face demarcated through following formula:

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

(3) detect dog nostril and two dog eyes simultaneously, but dog nostril and dog eyes have coincidence, then the testing result in dog nostril is a false-alarm, and the testing result of dog eyes is correct; Then said dog face is demarcated through following formula:

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=\min (y_{a1},y_{b2})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

(4) detect dog nostril and dog eyes simultaneously, but dog nostril and dog eyes have coincidence, then the testing result of dog nostril and dog eyes is false-alarm, can't demarcate the dog face.

(5) only detect one or two dog eyes, then the testing result of dog eyes is false-alarm, can't demarcate the dog face.

(6) do not detect whatever, then showing in the image to be detected does not have dog.

Need to prove,, after said dog face being demarcated, occur through the position in the detected dog of said dog nostril detecting device nostril and through the position of the detected dog eyes of said dog eye detector for above-mentioned testing process

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}\&le;0\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f\&le;0\end{array}\right.$ Then $\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=1\\ {\mathrm{<figure-callout\; id="1"\; label="y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_f=1\end{array}\right.$

$\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}\&GreaterEqual;\mathrm{<figure-callout\; id="1"\; label="W\; y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">W</figure-callout>}& \\ y_f{1}_{}\&GreaterEqual;H\end{array}\right.$ Then $\left\{\begin{array}{c}{x}_{\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">f</figure-callout>}1}=\mathrm{<figure-callout\; id="1"\; label="W\; y\; f"\; filenames="HDA0000131329090000021.png"\; state="\{\{state\}\}">W</figure-callout>}& \\ y_f{1}_{}=H\end{array}\right.$

Wherein, W, H are respectively the wide and high of image to be detected.

Table 3 is the dog face detecting methods that adopt present embodiment, the test result that obtains when image to be detected is detected:

Table 3

Dog total number of images (width of cloth)

565

Average resolution rate (pixel)	652×488
		Dog strains class sum (kind)	Greater than 80
Dog face sum (individual)	708
		Detect accuracy (%)	86.86％
Detect false alarm rate (individual/image)	0.152

See also Fig. 8, Fig. 8 is the actual example figure after adopting the dog face detecting method of the embodiment of the invention three to treat detected image to detect; The rectangle frame of wherein demarcating on the dog nostril is the result that dog nostril detecting device detects; The rectangle frame of demarcation on the dog eyes is the result that the dog eye detector detects, and maximum rectangle frame is the dog face testing result of demarcating according to the position relation of detected dog nostril and dog eyes.In the process that the dog face is demarcated, the testing result of dog eye detector has played booster action.Three dogs are different types of dog in the example illustrated, and image background is general life background, and the dog face is all accurately detected among the figure, and two cats in the middle instance also can correctly be distinguished, and overall testing result does not have false-alarm.For the image in the instance of the right, dog eyes do not detect, but this not influence with the dog nostril be that master, dog eyes are demarcated the dog face for assisting.The test result of associative list 3 and Fig. 8 can learn, through the dog face detecting method of present embodiment, can detect accurately the dog face.

Corresponding to above-mentioned dog face detecting method, the embodiment of the invention also provides a kind of dog face pick-up unit, sees also Fig. 9, and Fig. 9 is the structural representation of the dog face pick-up unit of the embodiment of the invention three, and is as shown in Figure 9, and said dog face pick-up unit comprises:

Dog nostril detecting device 101.

Dog eye detector 102.

Pretreatment unit 103 is suitable for treating detected image and carries out pre-service.

Dog nostril detecting unit 104 links to each other with said dog nostril detecting device 101, pretreatment unit 103, is suitable for adopting the dog nostril in 101 pairs of pretreated images to be detected of said dog nostril detecting device to detect.

Dog eye detection unit 105 links to each other with said dog eye detector 102, and the dog eyes that are suitable for adopting said dog eye detector 102 to treat in the detected image detect.

Demarcate unit 106, link to each other, be suitable for said dog face being demarcated based on the testing result of said dog nostril detecting unit 104 and said dog eye detection unit 105 with said dog nostril detecting unit 104, dog eye detection unit 105.

In the present embodiment, said pretreatment unit 103 comprises:

Segmentation unit for scaling (not shown) is suitable for based on the segmentation zoom factor said image to be detected being carried out convergent-divergent, and said segmentation zoom factor is associated with the size of said dog nostril detecting device.

The rotary unit (not shown) is suitable for the image rotation predetermined angular to be detected behind the convergent-divergent.

Left and right sides roll-over unit (not shown) is suitable for after said segmentation unit for scaling carries out convergent-divergent to said image to be detected, before the image rotation predetermined angular to be detected of said rotary unit after to convergent-divergent, and upset about the image to be detected behind the said convergent-divergent carried out.

Said dog nostril detecting unit 104 comprises:

The first cutting unit (not shown) is suitable for according to the size of current search window pretreated image to be detected being cut apart at least one search subwindow of acquisition.

The first detecting unit (not shown) is suitable for detecting said search subwindow through said dog nostril detecting device 101, until detecting the search subwindow that has said dog nostril.

The first updating block (not shown) is suitable in said first detection upgrading the size of current search window with first preset multiple when having the search subwindow in said dog nostril.

When the first control module (not shown), the size that is suitable for the current search window after said first updating block upgrades are less than or equal to said pretreated size of images to be detected, the work of control said units.

The first output unit (not shown) is suitable in said first detection when having the search subwindow in said dog nostril, and there is the position of the search subwindow in said dog nostril in output.

Said dog eye detection unit 105 comprises:

The second cutting unit (not shown) is suitable for treating detected image according to the size of current search window and cuts apart at least one search subwindow of acquisition.

The second detecting unit (not shown) is suitable for detecting said search subwindow through said dog eye detector 102, until detecting the search subwindow that has said dog eyes.

The second updating block (not shown) is suitable in said second detection upgrading the size of current search window with second preset multiple when having the search subwindow of said dog eyes.

When the second control module (not shown), the size that is suitable for the current search window after said second updating block upgrades are less than or equal to said size of images to be detected, the work of control said units.

The second output unit (not shown) is suitable in said second detection when having the search subwindow of said dog eyes, and there is the position of the search subwindow of said dog eyes in output.

In the present embodiment, said dog face pick-up unit can carry out referring to above-mentioned dog face detecting method the process that the dog face detects, and repeats no more here.

Embodiment four

In the present embodiment; Detect the dog face through adopting said dog nostril detecting device to detect the dog nostril or adopt said dog eye detector to detect the dog eyes; Promptly adopt dog face property detector to detect dog face characteristic, and then come the dog face is demarcated through said dog face characteristic.Below the method that adopts dog face property detector to detect the dog face is explained accordingly.

See also Figure 10, Figure 10 is the process flow diagram of the dog face detecting method of the embodiment of the invention four; Shown in figure 10, said dog face detecting method comprises:

Step S4 ': treat detected image and carry out pre-service.

Step S5 ': adopt dog face property detector that the dog face characteristic in the pretreated image to be detected is detected.

Step S6 ': the testing result through said dog face property detector is demarcated said dog face.

In the present embodiment; Said dog face property detector both can also can be the dog eye detector for dog nostril detecting device; And treat detected image in the present embodiment and carry out pre-service; And adopt dog nostril detecting device to the dog nostril in the pretreated image to be detected detect with embodiment three in similar; Adopt the dog eye detector that the dog eyes in the pretreated image to be detected are detected and to adopt dog nostril detecting device that the dog nostril in the pretreated image to be detected is detected similar, the demarcation of treating dog face in the detected image also with embodiment three in similar, so locate no longer to launch concrete detailed description.

Corresponding to above-mentioned dog face detecting method, the embodiment of the invention also provides a kind of dog face pick-up unit, sees also Figure 11, and Figure 11 is the structural representation of the dog face pick-up unit of the embodiment of the invention four, and is shown in figure 11, and said dog face pick-up unit comprises:

Dog face property detector 201.

Pretreatment unit 202 is suitable for treating detected image and carries out pre-service.

Detecting unit 203 links to each other with said dog face property detector 201, pretreatment unit 202, is suitable for adopting the dog face characteristic in 201 pairs of pretreated images to be detected of said dog face property detector to detect.

Demarcate unit 204, link to each other, be suitable for said dog face being demarcated based on the testing result of said detecting unit 203 with said detecting unit 203.

In the present embodiment, said pretreatment unit 202 comprises:

Segmentation unit for scaling (not shown) is suitable for based on the segmentation zoom factor said image to be detected being carried out convergent-divergent, and said segmentation zoom factor is associated with the size of dog face property detector.

The rotary unit (not shown) is suitable for the image rotation predetermined angular to be detected behind the convergent-divergent.

Left and right sides roll-over unit (not shown) is suitable for after said segmentation unit for scaling carries out convergent-divergent to said image to be detected, before the image rotation predetermined angular to be detected of said rotary unit after to convergent-divergent, and upset about the image to be detected behind the said convergent-divergent carried out.

Said detecting unit 203 comprises:

The cutting unit (not shown) is suitable for according to the size of current search window pretreated image to be detected being cut apart at least one search subwindow of acquisition.

The characteristic detection unit (not shown) is suitable for detecting said search subwindow through said dog face property detector 201, until detecting the search subwindow that has said dog face characteristic.

The updating block (not shown) is suitable for detecting when having the search subwindow of said dog face characteristic in said characteristic detection unit, upgrades the size of current search window with preset multiple.

When control module (not shown), the size of the current search window that is suitable for upgrading at said updating block are less than or equal to pretreated size of images to be detected, the work of control said units.

The output unit (not shown) is suitable for when said characteristic detection unit detects the search subwindow that has said dog face characteristic, exporting the position of the search subwindow of said dog face characteristic.

In the present embodiment, said dog face pick-up unit can carry out referring to above-mentioned dog face detecting method the process that the dog face detects, and repeats no more here.

In sum, the technical scheme of the embodiment of the invention has following beneficial effect at least:

Through with dog face characteristic image as the positive sample of Adaboost method, do not comprise the negative sample of the image of said dog face characteristic as the Adaboost method; And then train the performance of dog face property detector of acquisition good through the Adaboost method, can detect the characteristic of dog face accurately and rapidly.

Through adopting above-mentioned dog face property detector that the dog face in the pretreated image to be detected is detected, can detect the position at dog face place exactly, and the accuracy when said dog face detected is high, real-time is good.

Further; Said pre-service comprises that treating detected image carries out rotating predetermined angular again behind the segmentation convergent-divergent; Make when adopting said dog face property detector that the dog face is detected; Can detect the relatively poor image to be detected of picture quality; And when treating detected image and carrying out adopting said dog face property detector that the dog face is detected again after the above-mentioned pre-service, can also accelerate having the detection speed of high-resolution image to be detected, and then realize the dog face in the different images to be detected is carried out accurate detection.

Further; Through dog nostril detecting device the dog nostril in the pretreated image to be detected is detected, the dog eyes of treating in the detected image through the dog eye detector detect, and it is main having realized detecting with the dog nostril; The dog eye detection detects the dog face for assisting; Because be main with the dog nostril in the testing process, the dog eyes are auxilliary, so can remove the detection false-alarm in the dog nostril that possibly hit on eyes; And then can realize detection more accurately and rapidly to the dog face; And with the testing result of dog nostril detecting device is that the testing result of master, dog eye detector is auxilliary the dog face to be detected, and real-time accuracy good, that detect is high, and false alarm rate is very low.

Though the present invention with preferred embodiment openly as above; But it is not to be used for limiting the present invention; Any those skilled in the art are not breaking away from the spirit and scope of the present invention; Can utilize the method and the technology contents of above-mentioned announcement that technical scheme of the present invention is made possible change and modification, therefore, every content that does not break away from technical scheme of the present invention; To any simple modification, equivalent variations and modification that above embodiment did, all belong to the protection domain of technical scheme of the present invention according to technical spirit of the present invention.

Claims (39)

1. the formation method of a dog face property detector is characterized in that, comprises the steps:

Extract the dog face characteristic image of first predetermined quantity;

Extract the image of not comprising of second predetermined quantity of said dog face characteristic; Said second predetermined quantity is greater than said first predetermined quantity;

With the positive sample of said dog face characteristic image as the Adaboost method, the said image that does not comprise said dog face characteristic obtains said dog face property detector as the negative sample of Adaboost method.

2. the formation method of dog face property detector as claimed in claim 1 is characterized in that, extracts dog face characteristic image and comprises:

From training image, extract the image that comprises dog face characteristic;

The said image that comprises dog face characteristic is carried out normalization, obtain dog face characteristic image.

3. the formation method of dog face property detector as claimed in claim 2 is characterized in that, the said image that comprises dog face characteristic that from training image, extracts comprises:

Demarcate the upper left corner and the position/upper right corner, the lower right corner and the position, the lower left corner of dog face characteristic in the said training image;

The center that obtains dog face characteristic in the said training image based on the upper left corner of said dog face characteristic and position/upper right corner, the lower right corner and position, the lower left corner;

Center with said dog face characteristic is a symcenter; The upper left corner of moving said dog face characteristic and position/upper right corner, the lower right corner and position, the lower left corner are wide identical until the upper left corner of said dog face characteristic and the lower right corner/upper right corner and lower left corner distance in the horizontal direction and said dog face characteristic, high identical in the distance of vertical direction and said dog face characteristic.

4. the formation method of dog face property detector as claimed in claim 2 is characterized in that, the said image that comprises dog face characteristic that from training image, extracts comprises:

Foundation comprises the rectangular window of said dog face characteristic;

Adjust the size and the position of said rectangular window, circumscribed until the edge of adjusted rectangular window and said dog face characteristic.

5. the formation method of dog face property detector as claimed in claim 2 is characterized in that, saidly the said image that comprises dog face characteristic is carried out normalization comprises:

Normalization size and the said size of images that comprises dog face characteristic based on predetermined are obtained the convergent-divergent multiple;

Based on said convergent-divergent multiple, the said image that comprises dog face characteristic is carried out interpolation.

6. the formation method of dog face property detector as claimed in claim 2 is characterized in that, when the said image that comprises dog face characteristic is coloured image, the said gray level image that comprises the image of dog face characteristic is carried out normalization.

7. the formation method of dog face property detector as claimed in claim 2 is characterized in that, also comprises: upset about the image that comprises dog face characteristic after the normalization is carried out.

8. the formation method of dog face property detector as claimed in claim 7 is characterized in that, the said upset about the image that comprises dog face characteristic after the normalization is carried out obtains through following mode:

$\left\{\begin{array}{c}{P}_1(i,j)=P(i,n_c-j+1)\\ 1\&le;i\&le;n_r\\ 1\&le;j\&le;n_c\end{array}\right.$

Wherein, P (i, n _c-j+1) be the capable n of i in the image that comprises dog face characteristic after the normalization _cThe pixel value of-j+1 row, P ₁(i, j) be about the pixel value of the capable j of i row in the image after the upset, n _r* n _cIt is the size of images that comprises dog face characteristic after the normalization.

9. the formation method of dog face property detector as claimed in claim 1; It is characterized in that; Said with the positive sample of said dog face characteristic image as the Adaboost method, the said image that does not comprise said dog face characteristic obtains said dog face property detector as the negative sample of Adaboost method and comprises:

Training parameter in the said Adaboost method is set, and said training parameter comprises: positive sample number, negative sample number, just/negative sample normalization size, progression, the positive sample percent of pass of each grade minimum and the maximum false alarm rate of each grade;

Obtain said dog face property detector based on said training parameter, said positive sample, negative sample.

10. the formation method of dog face property detector as claimed in claim 9 is characterized in that, said dog face is characterized as the dog nostril, and said dog face property detector is a dog nostril detecting device, and the training parameter in the said Adaboost method is:

Said positive sample number is more than or equal to 500;

Said negative sample number is 1.5～5 times of said positive sample number;

Just said/negative sample normalization is of a size of 20 * 20～28 * 28 pixels;

Said progression is 15～25 grades;

The positive sample percent of pass of said each grade minimum is 0.995～0.998;

The maximum false alarm rate of said each grade is 0.46～0.50.

11. the formation method of dog face property detector as claimed in claim 9 is characterized in that, said dog face is characterized as dog eyes, and said dog face property detector is the dog eye detector, and the training parameter in the said Adaboost method is:

Said positive sample number is more than or equal to 400;

Said negative sample number is 2～4 times of said positive sample number;

Just said/negative sample normalization is of a size of 10 * 15～20 * 30 pixels;

Said progression is 15～25 grades;

The positive sample percent of pass of said each grade minimum is 0.995～0.998;

The maximum false alarm rate of said each grade is 0.46～0.50.

12. a dog face detecting method is characterized in that, comprising:

Treat detected image and carry out pre-service;

The dog face property detector that adopts claim 1 to obtain detects the dog face characteristic in the pretreated image to be detected;

Testing result through said dog face property detector is demarcated said dog face.

13. dog face detecting method as claimed in claim 12 is characterized in that, the said detected image of treating is carried out pre-service and is comprised:

Based on the segmentation zoom factor said image to be detected is carried out convergent-divergent, said segmentation zoom factor is associated with the size of dog face property detector;

To the image rotation predetermined angular to be detected behind the convergent-divergent.

14. dog face detecting method as claimed in claim 13 is characterized in that, the said detected image of treating is carried out pre-service and is also comprised:

After said image to be detected is carried out convergent-divergent, before the rotation of the image to be detected behind convergent-divergent predetermined angular, upset about the image to be detected behind the said convergent-divergent carried out.

15. dog face detecting method as claimed in claim 12 is characterized in that, the dog face property detector that said employing claim 1 obtains detects the dog face characteristic in the pretreated image to be detected and comprises:

Size according to the current search window is cut apart at least one search subwindow of acquisition to pretreated image to be detected;

Detect said search subwindow through said dog face property detector, until detecting the search subwindow that has said dog face characteristic;

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than pretreated size of images to be detected;

There is the position of the search subwindow of said dog face characteristic in output.

16. dog face detecting method as claimed in claim 12; It is characterized in that; Said dog face is characterized as the dog nostril; Said dog face property detector is a dog nostril detecting device, and said testing result through said dog face property detector is demarcated said dog face and comprised through following formula and demarcate said dog face:

$\left\{\begin{array}{c}{x}_{f1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ y_{f1}=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

Wherein, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

17. dog face detecting method as claimed in claim 12; It is characterized in that; Said dog face is characterized as dog eyes; Said dog face property detector is the dog eye detector, and said testing result through said dog face property detector is demarcated said dog face and comprised through following formula and demarcate said dog face:

$\left\{\begin{array}{c}{x}_{f1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ y_{f1}=\min (y_{a1},y_{b1})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

Wherein, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

18. a dog face detecting method is characterized in that, comprising:

Treat detected image and carry out pre-service;

The dog nostril detecting device that adopts claim 10 to obtain detects the dog nostril in the pretreated image to be detected;

The dog eyes that the dog eye detector that adopts claim 11 to obtain is treated in the detected image detect;

Testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face.

19. dog face detecting method as claimed in claim 18 is characterized in that, the said detected image of treating is carried out pre-service and is comprised:

Based on the segmentation zoom factor said image to be detected is carried out convergent-divergent, said segmentation zoom factor is associated with the size of dog nostril detecting device;

To the image rotation predetermined angular to be detected behind the convergent-divergent.

20. dog face detecting method as claimed in claim 19 is characterized in that, the said detected image of treating is carried out pre-service and is also comprised:

After said image to be detected is carried out convergent-divergent, before the rotation of the image to be detected behind convergent-divergent predetermined angular, upset about the image to be detected behind the convergent-divergent carried out.

21. dog face detecting method as claimed in claim 18 is characterized in that, the dog nostril detecting device that said employing claim 10 obtains detects the dog nostril in the pretreated image to be detected and comprises:

Size according to the current search window is cut apart at least one search subwindow of acquisition to pretreated image to be detected;

Detect said search subwindow through said dog nostril detecting device, until detecting the search subwindow that has said dog nostril;

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than pretreated size of images to be detected;

There is the position of the search subwindow in said dog nostril in output.

22. dog face detecting method as claimed in claim 18 is characterized in that, the dog eye detector that said employing claim 11 obtains is treated dog eyes in the detected image and is detected and comprise:

Treat detected image according to the size of current search window and cut apart at least one search subwindow of acquisition;

Detect said search subwindow through said dog eye detector, until detecting the search subwindow that has said dog eyes;

If do not exist, then upgrade the size of current search window with preset multiple, repeat said process, until the size of current search window greater than size of images to be detected;

There is the position of the search subwindow of said dog eyes in output.

23. dog face detecting method as claimed in claim 18 is characterized in that, said testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face and is comprised:

When detecting dog nostril and two dog eyes, and the dog nostril do not overlap with two dog eyes, then through following formula said dog face demarcated:

$\left\{\begin{array}{c}{x}_a=0.5\&times;(x_{a2}+x_{a1})\\ y_a=0.5\&times;(y_{a2}+y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_b=0.5\&times;(x_{b2}+x_{b1})\\ y_b=0.5\&times;(y_{b2}+y_{b1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_c=0.5\&times;(x_{c2}+x_{c1})\\ y_c=0.5\&times;(y_{c2}+y_{c1})\end{array}\right.$

$\left\{\begin{array}{c}{x}_{f1}=x_a-0.5\&times;(x_b-x_a)\\ y_{f1}=\min (y_a,y_b)-0.25\&times;(y_c-\min (y_a,y_b))\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_b+0.5\&times;(x_b-x_a)\\ y_{f2}=y_c+0.5\&times;(y_c-\min (y_a,y_b))\end{array}\right.$

Wherein, (x _a, y _a) be the center of the left eye eyeball of calibrated dog, (x _b, y _b) be the center of the right eye eyeball of calibrated dog, (x _c, y _c) be the center in calibrated dog nostril, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

24. dog face detecting method as claimed in claim 18 is characterized in that, said testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face and is comprised:

Do not overlap with dog eyes when detecting dog nostril and dog eyes and said dog nostril; Perhaps only detect the dog nostril, then said dog face demarcated through following formula:

$\left\{\begin{array}{c}{x}_{f1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ y_{f1}=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

Wherein, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

25. dog face detecting method as claimed in claim 18 is characterized in that, said testing result through said dog nostril detecting device and said dog eye detector is demarcated said dog face and is comprised:

When detecting dog nostril and two dog eyes, and said dog nostril overlaps with dog eyes, then through following formula said dog face demarcated:

$\left\{\begin{array}{c}{x}_{f1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ y_{f1}=\min (y_{a1},y_{b1})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

Wherein, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

26. a dog face pick-up unit is characterized in that, comprising:

The dog face property detector that claim 1 obtains;

Pretreatment unit is suitable for treating detected image and carries out pre-service;

Detecting unit is suitable for adopting said dog face property detector that the dog face characteristic in the pretreated image to be detected is detected;

Demarcate the unit, be suitable for said dog face being demarcated based on the testing result of said detecting unit.

27. dog face pick-up unit as claimed in claim 26 is characterized in that said pretreatment unit comprises:

The segmentation unit for scaling is suitable for based on the segmentation zoom factor said image to be detected being carried out convergent-divergent, and said segmentation zoom factor is associated with the size of dog face property detector:

Rotary unit is suitable for the image rotation predetermined angular to be detected behind the convergent-divergent.

28. dog face pick-up unit as claimed in claim 27 is characterized in that said pretreatment unit also comprises:

Left and right sides roll-over unit is suitable for after said segmentation unit for scaling carries out convergent-divergent to said image to be detected, before the image rotation predetermined angular to be detected of said rotary unit after to convergent-divergent, and upset about the image to be detected behind the said convergent-divergent carried out.

29. dog face pick-up unit as claimed in claim 26 is characterized in that said detecting unit comprises:

Cutting unit is suitable for according to the size of current search window pretreated image to be detected being cut apart at least one search subwindow of acquisition;

Characteristic detection unit is suitable for detecting said search subwindow through said dog face property detector, until detecting the search subwindow that has said dog face characteristic;

Updating block is suitable for detecting when having the search subwindow of said dog face characteristic in said characteristic detection unit, upgrades the size of current search window with preset multiple;

When control module, the size of the current search window that is suitable for upgrading at said updating block are less than or equal to pretreated size of images to be detected, the work of control said units;

Output unit is suitable for when said characteristic detection unit detects the search subwindow that has said dog face characteristic, exporting the position of the search subwindow of said dog face characteristic.

30. dog face pick-up unit as claimed in claim 26 is characterized in that, said dog face property detector is a dog nostril detecting device, and said dog face is demarcated through following formula in said demarcation unit:

$\left\{\begin{array}{c}{x}_{f1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ y_{f1}=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

Wherein, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

31. dog face pick-up unit as claimed in claim 26 is characterized in that, said dog face property detector is the dog eye detector, and said dog face is demarcated through following formula in said demarcation unit:

$\left\{\begin{array}{c}{x}_{f1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ y_{f1}=\min (y_{a1},y_{b1})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

Wherein, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

32. a dog face pick-up unit is characterized in that, comprising:

The dog nostril detecting device that claim 10 obtains;

The dog eye detector that claim 11 obtains;

Pretreatment unit is suitable for treating detected image and carries out pre-service;

Dog nostril detecting unit is suitable for adopting said dog nostril detecting device that the dog nostril in the pretreated image to be detected is detected;

Dog eye detection unit, the dog eyes that are suitable for adopting said dog eye detector to treat in the detected image detect;

Demarcate the unit, be suitable for said dog face being demarcated based on the testing result of said dog nostril detecting unit and said dog eye detection unit.

33. dog face pick-up unit as claimed in claim 32 is characterized in that said pretreatment unit comprises:

The segmentation unit for scaling is suitable for based on the segmentation zoom factor said image to be detected being carried out convergent-divergent, and said segmentation zoom factor is associated with the size of said dog nostril detecting device:

Rotary unit is suitable for the image rotation predetermined angular to be detected behind the convergent-divergent.

34. dog face pick-up unit as claimed in claim 33 is characterized in that said pretreatment unit also comprises:

Left and right sides roll-over unit is suitable for after said segmentation unit for scaling carries out convergent-divergent to said image to be detected, before the image rotation predetermined angular to be detected of said rotary unit after to convergent-divergent, and upset about the image to be detected behind the said convergent-divergent carried out.

35. dog face pick-up unit as claimed in claim 32 is characterized in that, said dog nostril detecting unit comprises:

First cutting unit is suitable for according to the size of current search window pretreated image to be detected being cut apart at least one search subwindow of acquisition;

First detecting unit is suitable for detecting said search subwindow through said dog nostril detecting device, until detecting the search subwindow that has said dog nostril;

First updating block is suitable in said first detection upgrading the size of current search window with first preset multiple when having the search subwindow in said dog nostril;

When first control module, the size that is suitable for the current search window after said first updating block upgrades are less than or equal to said pretreated size of images to be detected, the work of control said units;

First output unit is suitable in said first detection when having the search subwindow in said dog nostril, and there is the position of the search subwindow in said dog nostril in output.

36. dog face pick-up unit as claimed in claim 32 is characterized in that, said dog eye detection unit comprises:

Second cutting unit is suitable for treating detected image according to the size of current search window and cuts apart at least one search subwindow of acquisition;

Second detecting unit is suitable for detecting said search subwindow through said dog eye detector, until detecting the search subwindow that has said dog eyes;

Second updating block is suitable in said second detection upgrading the size of current search window with second preset multiple when having the search subwindow of said dog eyes;

When second control module, the size that is suitable for the current search window after said second updating block upgrades are less than or equal to said size of images to be detected, the work of control said units;

Second output unit is suitable in said second detection when having the search subwindow of said dog eyes, and there is the position of the search subwindow of said dog eyes in output.

37. dog face pick-up unit as claimed in claim 32; It is characterized in that; Said dog nostril detecting unit and said dog eye detection unit detect dog nostril and two dog eyes; And said dog nostril does not overlap with two dog eyes, and said demarcation unit is demarcated said dog face through following formula:

$\left\{\begin{array}{c}{x}_a=0.5\&times;(x_{a2}+x_{a1})\\ y_a=0.5\&times;(y_{a2}+y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_b=0.5\&times;(x_{b2}+x_{b1})\\ y_b=0.5\&times;(y_{b2}+y_{b1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_c=0.5\&times;(x_{c2}+x_{c1})\\ y_c=0.5\&times;(y_{c2}+y_{c1})\end{array}\right.$

$\left\{\begin{array}{c}{x}_{f1}=x_a-0.5\&times;(x_b-x_a)\\ y_{f1}=\min (y_a,y_b)-0.25\&times;(y_c-\min (y_a,y_b))\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_b+0.5\&times;(x_b-x_a)\\ y_{f2}=y_c+0.5\&times;(y_c-\min (y_a,y_b))\end{array}\right.$

Wherein, (x _a, y _a) be the center of the left eye eyeball of calibrated dog, (x _b, y _b) be the center of the right eye eyeball of calibrated dog, (x _c, y _c) be the center in calibrated dog nostril, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

38. dog face pick-up unit as claimed in claim 32 is characterized in that, said dog nostril detecting unit and said dog eye detection unit detect dog nostril and dog eyes, and said dog nostril does not overlap with dog eyes; Perhaps only detect the dog nostril, said demarcation unit is demarcated said dog face through following formula:

$\left\{\begin{array}{c}{x}_{f1}=x_{c1}-1.25\&times;(x_{c2}-x_{c1})\\ y_{f1}=y_{c1}-2.5\&times;(y_{c2}-y_{c1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{c2}+1.25\&times;(x_{c2}-x_{c1})\\ y_{f2}=y_{c2}+0.5\&times;(y_{c2}-y_{c1})\end{array}\right.$

Wherein, (x _C1, y _C1) and (x _C2, y _C2) be respectively the position in the upper left corner, detected dog nostril and the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

39. dog face pick-up unit as claimed in claim 32; It is characterized in that; Said dog nostril detecting unit and said dog eye detection unit detect dog nostril and two dog eyes; And said dog nostril overlaps with dog eyes, and said demarcation unit is demarcated said dog face through following formula:

$\left\{\begin{array}{c}{x}_{f1}=x_{a1}-0.5\&times;(x_{a2}-x_{a1})\\ y_{f1}=\min (y_{a1},y_{b1})-0.5\&times;(y_{a2}-y_{a1})\end{array}\right.$ $\left\{\begin{array}{c}{x}_{f2}=x_{b2}+0.5\&times;(x_{b2}-x_{b1})\\ y_{f2}=\max (y_{a2},y_{b2})+0.5\&times;(y_{b2}-y_{b1})\end{array}\right.$

Wherein, (x _A1, y _A1) and (x _A2, y _A2) be respectively the left eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _B1, y _B1) and (x _B2, y _B2) be respectively the right eye eyeball upper left corner of detected dog and the position in the lower right corner, (x _F1, y _F1) and (x _F2, y _F2) be respectively the dog face upper left corner of demarcation and the position in the lower right corner.

Images