Summary of the invention
The problem that the present invention solves is to provide all low method for detecting human face and devices of a kind of complexity, false drop rate and loss.
In order to address the above problem, the invention provides a kind of method for detecting human face, comprising:
Utilize people's face detection algorithm to obtain a plurality of regional frames in image;
Described a plurality of regional frames are carried out cluster analysis, to obtain at least one cluster areas;
Add up the number of the regional frame in each cluster areas, the number of definite area frame is human face region greater than the cluster areas of first threshold.
Optionally, described method for detecting human face also comprises:
Determine human face region based on the number of regional frame less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold.
Optionally, described number based on regional frame determines that less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold human face region comprises:
Utilize the Face Detection algorithm to detect the skin pixel point of described cluster areas;
The number of described skin pixel point than described cluster areas in the value of number of pixel during greater than the 3rd threshold value, determine that described cluster areas is human face region.
Optionally, described number based on regional frame determines that less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold human face region comprises:
Utilize the Face Detection algorithm to detect the skin pixel point of arbitrary regional frame in described cluster areas;
The number of described skin pixel point than described arbitrary regional frame in the value of number of pixel during greater than the 4th threshold value, determine that described cluster areas is human face region.
Optionally, described number based on regional frame determines that less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold human face region comprises:
Utilize the Face Detection algorithm to detect the skin pixel point of the predeterminable area of arbitrary regional frame in described cluster areas;
During greater than the 5th threshold value, determine that described cluster areas is human face region than the value of the number of the pixel in described predeterminable area in the number of described skin pixel point.
Optionally, described regional frame is rectangle frame, and described predeterminable area is determined by following manner:
width∈0.25W~1W,height∈0.25H~1H;
Wherein, width is the wide of described predeterminable area, and height is the height of described predeterminable area, and W is the wide of described rectangle frame, and H is the height of described rectangle frame.
Optionally, described method for detecting human face also comprises: the number of definite area frame is non-face zone less than the cluster areas of Second Threshold.
Optionally, described first threshold is the integer between [10,40].
Optionally, described Second Threshold is the integer between [5,8].
Optionally, the span of described the 3rd threshold value is [0.3~0.8].
Optionally, described Face Detection algorithm is the Face Detection algorithm based on Gauss model or Bayesian model.
Optionally, described method for detecting human face also comprises: before the number of the regional frame in each cluster areas of statistics, remove in described cluster areas and meet pre-conditioned regional frame.
Optionally, describedly pre-conditionedly be: at the center of described regional frame to the distance at the center of cluster areas than the value of the width of described regional frame less than distance threshold.
Optionally, described regional frame is rectangle frame, and the scope of described distance threshold is [0.5W, W], and wherein W is the wide of described rectangle frame.
Optionally, described people's face detection algorithm and Face Detection algorithm carry out at different color spaces.
Optionally, described people's face detection algorithm is the adaboost algorithm.
For addressing the above problem, the present invention also provides a kind of people's face pick-up unit, comprising:
Acquiring unit, be used for utilizing people's face detection algorithm to obtain a plurality of regional frames of image;
The cluster analysis unit, be used for described a plurality of regional frames are carried out cluster analysis, to obtain at least one cluster areas;
Statistic unit, for the number of the regional frame of adding up each cluster areas;
The first determining unit, the number that is used for the definite area frame is human face region greater than the cluster areas of first threshold.
Compared with prior art, technical scheme of the present invention has the following advantages:
A plurality of regional frames that obtain by people's face detection algorithm are carried out cluster analysis, add up the number of regional frame in each cluster areas, with this, detect human face region, for the method for detecting human face of skin color based, because it is subjected to illumination effect less, therefore loss is low; For the method for detecting human face based on Adaboost, owing to need not to gather a large amount of samples in the training process of detecting device, therefore reduced the complexity that people's face detects; And, after adopting people's face detection algorithm to obtain regional frame, first carry out cluster analysis, then the regional frame number in cluster areas is added up, also reduced the false drop rate that people's face detects.
In the number of the regional frame of cluster areas during less than first threshold and greater than Second Threshold, features of skin colors based on described cluster areas is determined human face region, compared to the method for detecting human face of skin color based with based on people's face detection algorithm of Adaboost, carry out people's face by the statistics of the regional frame number in conjunction with cluster areas and Face Detection algorithm and detect, further reduced loss and false drop rate that people's face detects.
Embodiment
, for above-mentioned purpose of the present invention, feature and advantage can more be become apparent, below in conjunction with accompanying drawing, the specific embodiment of the present invention is described in detail.
Set forth detail in the following description so that fully understand the present invention.But the present invention can be different from alternate manner described here and implements with multiple, and those skilled in the art can be in the situation that do similar popularization without prejudice to intension of the present invention.Therefore the present invention is not subjected to the restriction of following public embodiment.
People's face detection algorithm of existing skin color based easily is subject to illumination effect, therefore lower to the verification and measurement ratio of people's face, although and based on the method for detecting human face of Adaboost, can obtain by strict training the strong classifier of cascade to detect exactly people's face, but during the training strong classifier, computation complexity is higher, is not suitable for the image in mini-plant is detected.
The inventor considers whether can carry out accurately and comprehensively detecting to people's face under the prerequisite that does not increase computation complexity, the inventor finds through long-term research, when adopting people's face detection algorithm detected image, can detect a plurality of regional frames, after a plurality of regional frames are carried out cluster analysis, can obtain different cluster areas, and in each cluster areas, the detection of the number of regional frame and human face region there is close relationship.
Therefore, the inventor proposes, and the judgement of the number by the regional frame to cluster areas detects human face region.
Embodiment one
Fig. 1 is the schematic flow sheet of the method for detecting human face of the embodiment of the present invention one, and as shown in Figure 1, described method for detecting human face comprises:
Step S11: utilize people's face detection algorithm to obtain a plurality of regional frames in image;
Step S12: described a plurality of regional frames are carried out cluster analysis, to obtain at least one cluster areas;
Step S13: add up the number of the regional frame in each cluster areas, the number of definite area frame is human face region greater than the cluster areas of first threshold.
Execution step S11, can apply existing human face detection tech in this step and obtain a plurality of regional frames in image, as: can adopt algorithm based on bayesian criterion, based on the algorithm of artificial neural network, algorithm or the Adaboost algorithm of support vector machine, image be detected, obtain a plurality of regional frames in image.In the present embodiment, to adopt the Adaboost algorithm, image is detected as example and describes.
Described employing Adaboost algorithm detects actual referring to image and forms strong classifier by the Adaboost algorithm, utilizes described strong classifier to detect image.In general, can, by the positive sample that comprises people's face that gathers some and the negative sample that does not comprise people's face, form the sample training collection.Practice Weak Classifier for different sample training training, then these Weak Classifiers that obtain on different training sets are gathered, form a strong classifier.Described different sample training collection is to realize by adjusting weight corresponding to each sample.
While starting to train at first, the weight that each sample (comprising positive sample and negative sample) is corresponding is identical, trains a Weak Classifier h under this sample distribution
1(x).For h
1(x) sample of wrong minute, increase its corresponding weight; And for h
1(x) sample of correct classification, reduce its weight.Make the sample of wrong minute highlight, and obtain new sample distribution.The situation of according to mistake, dividing simultaneously, is given h
1(x) weight, represent that the significance level of this Weak Classifier, mistake are got fewer weight larger.Under new sample distribution, again to Weak Classifier h
1(x) train, obtain Weak Classifier h
2(x) and weight.The like,, through T circulation, just obtained T Weak Classifier, and T the weight that Weak Classifier is corresponding.Finally this T Weak Classifier is added up and just obtained strong classifier by its corresponding weight.Form strong classifier and can form the detailed process of strong classifier with reference to the Adaboost algorithm that utilizes of prior art.
In this step, while adopting Adaboost Algorithm for Training strong classifier, in order to reduce computation complexity, in training process, the extraction of the quantity of training sample, training parameter can be carried out suitable choosing.After obtaining strong classifier by the Adaboost Algorithm for Training, utilize described strong classifier to detect image.
In addition, need to prove, while adopting the Adaboost algorithm to detect people's face, carry out at gray space, therefore, if image to be detected is coloured image, before obtaining regional frame in image to be detected by the Adaboost algorithm, should be first with image transitions to be detected to gray space.
For the strong classifier that adopts the Adaboost algorithm to form, strong classifier during detection (detecting device) is generally rectangular window, and detecting device is to detect according to the pixel level line slip in image to be detected.Therefore, can detect a plurality of rectangle frames (the rectangle frame size is indefinite) in the place that people's face is arranged.
What Fig. 2 was that the employing Adaboost algorithm of the embodiment of the present invention one detects may be the distribution schematic diagram of human face region, for convenience of explanation, has only provided a corresponding regional frame of people's face in Fig. 2.For a plurality of people's faces in piece image, each people's face place all should occur and similar situation shown in Figure 2.
As shown in Figure 2, a plurality of rectangle frames that detecting device detects are in fact corresponding people's face, also can say these a plurality of rectangle frames in fact corresponding be the same area.Therefore, need to carry out cluster analysis to this situation merges.The rectangle frame that detects is merged it by clustering algorithm, to obtain corresponding cluster areas.For example: the rectangle frame that detects can be merged according to its center, the rectangle frame of namely center being assembled is merged into a rectangular area, described rectangular area is carries out the cluster areas that obtains after cluster analysis, the length of described cluster areas and wide can being taken at are respectively carried out in merging process rectangle frame, the maximal value of the maximal value of rectangle frame length and rectangle frame width, the center of described rectangular area is the center of cluster areas.In actual applications, specifically adopt which kind of clustering algorithm that a plurality of rectangle frames are carried out cluster analysis and merge to form cluster areas, by actual demand, determined.
Execution step S12, carry out cluster analysis to a plurality of rectangle frames that obtain in step S11, obtains N cluster areas (N 〉=1), is also N candidate face zone.And for N cluster areas, it all carries out cluster analysis by at least one rectangle frame and merges acquisition, has all comprised in other words at least one rectangle frame for each cluster areas.
In the present embodiment,, in order to prevent from cluster process occurring deviation, also comprise that the rectangle frame to carrying out cluster analysis detects, in other words the rectangle frame in cluster areas is detected, to remove ineligible rectangle frame.Particularly, the center of calculating exactly rectangle frame to the center c of cluster areas apart from d (referring to Fig. 2), when described d does not meet following formula, this rectangle frame is picked out cluster areas.
Wherein, W
iFor the width of i rectangle frame in cluster areas, rec_ratio is distance threshold.In the present embodiment, rec_ratio gets 0.5W
i~W
iBetween.
Execution step S13, add up the rectangle frame in N cluster areas obtaining in step S12 (candidate face zone), and the number by the rectangle frame added up detects human face region.The inventor has carried out a large amount of experiments, gather a large amount of images and performed step respectively S11 and S12, find when the cluster areas that obtains is human face region, the number of the rectangle frame of this cluster areas meets some requirements, and has obtained to detect according to the number of the rectangle frame in Statistical Clustering Analysis zone the fiducial interval of people's face.Particularly, be exactly the number of rectangle frame of cluster areas during greater than first threshold, the cluster areas of acquisition is human face region.Described first threshold is associated with the target detection rate of people's face detection algorithm, and described target detection rate comprises: the false drop rate of the accuracy of target detection, the loss of target detection and target detection.
Fig. 3 is the schematic diagram that concerns between the picture number of the embodiment of the present invention one and rectangle frame.As shown in Figure 3, in figure, horizontal ordinate has represented the number of the rectangle frame of cluster areas, and ordinate has represented the number of the image that gathers.As shown in Figure 3, in the present embodiment, in the image that collects, while human face region being detected, the number of rectangle frame is all greater than 1.In the present embodiment in other words, not there will be the situation of only having a rectangle frame at human face region.When cluster areas was human face region, in cluster areas, the number of rectangle frame was that the image of 2 has 139 width; When cluster areas was human face region, in cluster areas, the number of rectangle frame was that the image of 4 has 78 width; When cluster areas was human face region, in cluster areas, the number of rectangle frame was that the image of 15 has 162 width; ...; When cluster areas was human face region, in cluster areas, the number of rectangle frame was that the image of 20 has 173 width; ...; When cluster areas was human face region, in cluster areas, the number of rectangle frame was that the image of 25 has 186 width; When cluster areas was human face region, in cluster areas, the number of rectangle frame was that the image of 30 has 151 width; ...; During in high fiducial interval shown in Figure 3, described cluster areas is human face region when the number of the rectangle frame of the cluster areas in the image that collects.Described first threshold is the integer between [10,40], and in the present embodiment, described first threshold gets 10.Number by the rectangle frame to cluster areas is added up, and can detect soon the position of people's face.
Corresponding to above-mentioned method for detecting human face, the present embodiment also provides a kind of people's face pick-up unit, and Fig. 4 is the structural representation of people's face pick-up unit of the embodiment of the present invention one, and as shown in Figure 4, described people's face pick-up unit comprises:
Acquiring unit 101, be used for utilizing people's face detection algorithm to obtain a plurality of regional frames of image.
Cluster analysis unit 102, be connected with described acquiring unit 101, is used for described a plurality of regional frames are carried out cluster analysis, to obtain at least one cluster areas.
Statistic unit 103, be connected with described cluster analysis unit 102, is used for the number of the regional frame of each cluster areas of statistics.
The first determining unit 104, be connected with described statistic unit 103, and the number that is used for the definite area frame is human face region greater than the cluster areas of first threshold.
In the present embodiment, described people's face detection algorithm can be the Adaboost algorithm, and described first threshold is the integer between [10~40].
In the present embodiment, described people's face pick-up unit also comprises: remove the unit (not shown), before the number for the regional frame in described statistic unit 103 each cluster areas of statistics, remove in described cluster areas and meet pre-conditioned regional frame.Describedly pre-conditionedly be: at the center of described regional frame to the distance at the center of cluster areas than the value of the width of described regional frame less than distance threshold.Regional frame described in the present embodiment is rectangle frame, and the scope of described distance threshold is [0.5W, W], and wherein W is the wide of described rectangle frame.
In the present embodiment, the course of work of described people's face pick-up unit can be carried out referring to above-mentioned method for detecting human face, repeats no more herein.
Embodiment two
In embodiment one, while having provided number at the regional frame of cluster areas greater than first threshold, determine that this cluster areas is human face region.In the testing process of reality, the satisfied situation greater than first threshold of number that the regional frame of cluster areas also may occur, the present embodiment is during to the situation in the discontented sufficient embodiment one of the number of the regional frame of cluster areas, and whether the how to confirm cluster areas is that human face region describes.
Fig. 5 is the schematic flow sheet of the method for detecting human face of the embodiment of the present invention two, and as shown in Figure 5, described method for detecting human face comprises:
Step S 11: utilize people's face detection algorithm to obtain a plurality of regional frames in image;
Step S12: described a plurality of regional frames are carried out cluster analysis, to obtain at least one cluster areas;
Step S13: add up the number of the regional frame in each cluster areas, the number of definite area frame is human face region greater than the cluster areas of first threshold;
Step S14: based on the number of regional frame, less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold, determine human face region;
Step S15: the number of definite area frame is non-face zone less than the cluster areas of Second Threshold.
Similar in step S11~S13 and embodiment one in the present embodiment, so locate no longer to describe in detail.
Please continue referring to Fig. 3, between the zone of transition shown in Fig. 3, the number that is regional frame in the cluster areas that detects is less than first threshold and greater than the situation of Second Threshold, described Second Threshold is associated with the target detection rate of people's face detection algorithm, described Second Threshold is the integer between [5~8], in the present embodiment, described Second Threshold gets 5.For the number of the rectangle frame of cluster areas is in cluster areas between zone of transition, utilize the features of skin colors of this cluster areas to determine human face region.
Particularly, step S14 comprises:
Step S141: utilize the Face Detection algorithm to detect the skin pixel point of the predeterminable area of arbitrary regional frame in described cluster areas;
Step S142: in the number of described skin pixel point, than the value of the number of the pixel in described predeterminable area, during greater than the 5th threshold value, determine that described cluster areas is human face region.
Execution step S141, due to the cluster areas correspondence after cluster analysis at least one rectangle frame, therefore, utilize the Face Detection algorithm to detect the skin pixel point of the predeterminable area of arbitrary rectangle frame in described cluster areas.
Fig. 6 be the embodiment of the present invention two predeterminable area choose schematic diagram, described predeterminable area is rectangle, the wide width of described predeterminable area is between 0.25W~1W, W is wide (unit is pixel) of rectangle frame, the high height of described predeterminable area is between 0.25H~1H, and H is the height (unit is pixel) of rectangle frame.
In general, for the human face region structure of rectangle, vertically it equally spacedly can be divided into three zones, wherein eyes and eyebrow are usually located at 1/3 top zone of rectangular area; Face is usually located at 1/3 zone on the lower, rectangular area, and nose is positioned at remaining 1/3 zone that mediates.Detect human face region owing to being based on features of skin colors in the present embodiment, and the features of skin colors of eyes, eyebrow and face region is not obvious, therefore, for human face region to be detected (rectangular area), described predeterminable area choose the impact that should as far as possible avoid eyes eyebrow and face, choose the zone line (containing nosed zone) of described rectangular area.In the present embodiment, the choosing as shown in Figure 6 of described predeterminable area, the center that is centered close to rectangle frame of described predeterminable area, described predeterminable area wide
The height of described predeterminable area
After selected predeterminable area, can apply the skin pixel point of the described predeterminable area of existing Face Detection technology for detection, as: the skin pixel point that can adopt predeterminable area as described in detecting based on the Face Detection algorithm of Gauss model, also can adopt the skin pixel point that detects described predeterminable area based on the Face Detection algorithm of Bayesian model, and while adopting the Face Detection algorithm to detect skin pixel point, the space at image place can be any one in other color spaces such as YCbCr space, yuv space, YIQ space.The concrete skin pixel point that detects described predeterminable area based on the Face Detection algorithm of which kind of model that adopts, determined by actual conditions.
Execution step S142, the number of statistics skin pixel point, during greater than the 5th threshold value, determine that described cluster areas is human face region than the value of the number of all pixels in predeterminable area in the number of the skin pixel point of predeterminable area.That is:
Wherein, count is the number of the skin pixel point of predeterminable area, and weight * height is the number of the pixel in predeterminable area, and limit_ratio is the 5th threshold value.
Described the 5th threshold value is determined by test, the scope of the 5th threshold value described in the present embodiment be [0.5~1), can get 0.5,0.6,0.7 etc. as, described the 5th threshold value.
In another embodiment, determine that less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold human face region is by detecting the skin pixel point of cluster areas based on the number of regional frame, and the skin pixel point in cluster areas and the pixel in whole cluster areas are compared to carry out, particularly, comprising:
Utilize the Face Detection algorithm to detect the skin pixel point of described cluster areas;
The number of described skin pixel point than described cluster areas in the value of number of pixel during greater than the 3rd threshold value, determine that described cluster areas is human face region.
Described the 3rd threshold value is determined by test, and in the present embodiment, the span of described the 3rd threshold value is [0.3,0.8].Can get 0.3,0.4,0.5 etc. as, described the 3rd threshold value.
In another embodiment, determine that less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold human face region is by detecting the skin pixel point number of the arbitrary regional frame in described cluster areas based on the number of regional frame, and the pixel number in described skin pixel point number and this regional frame is compared to carry out, particularly, comprising:
Utilize the Face Detection algorithm to detect the skin pixel point of arbitrary regional frame in described cluster areas;
The number of described skin pixel point than described arbitrary regional frame in the value of number of pixel during greater than the 4th threshold value, determine that described cluster areas is human face region.
Described the 4th threshold value is determined by test, in the present embodiment, the span of described the 4th threshold value be [0.5~1), can get 0.5,0.6,0.7,0.8 etc. as, described the 4th threshold value.
, for the number of regional frame in the cluster areas that the detects situation less than Second Threshold, when namely the number of the rectangle frame shown in Fig. 3 is positioned at low fiducial interval, determine that this cluster areas is non-face zone.
Need to prove, in the present embodiment, be the number by the rectangle frame in Statistical Clustering Analysis zone, and the number of rectangle frame is classified and obtained some fiducial intervals.Yet Fig. 3 has only provided a kind of dividing mode of fiducial interval, namely according to the number of rectangle frame, marks off three fiducial intervals, when the number of the rectangle frame of adding up a certain cluster areas belongs to high fiducial interval, determines that this cluster areas is human face region.When the number of the rectangle frame of adding up a certain cluster areas belongs to low fiducial interval, determine that this cluster areas is non-face zone.When the number of the rectangle frame of adding up a certain cluster areas belongs between zone of transition, to determine whether this cluster areas is human face region further combined with the Face Detection algorithm.When practice, the division of fiducial interval not only is confined to shown in Figure 3, as, fiducial interval, can be divided into 2,4,5 etc., specifically how to divide rational fiducial interval and can test and determine by the data sample to collecting.Therefore, the division of fiducial interval should be as the restriction to technical solution of the present invention.
Corresponding to above-mentioned method for detecting human face, the embodiment of the present invention also provides a kind of people's face pick-up unit, and Fig. 7 is the structural representation of people's face pick-up unit of the embodiment of the present invention two, and as shown in Figure 7, described people's face pick-up unit comprises:
Acquiring unit 101, be used for utilizing people's face detection algorithm to obtain a plurality of regional frames of image.
Cluster analysis unit 102, be connected with described acquiring unit 101, is used for described a plurality of regional frames are carried out cluster analysis, to obtain at least one cluster areas.
Statistic unit 103, be connected with described cluster analysis unit 102, is used for the number of the regional frame of each cluster areas of statistics.
The first determining unit 104, be connected with described statistic unit 103, and the number that is used for the definite area frame is human face region greater than the cluster areas of first threshold.
The second determining unit 105, be connected with described statistic unit 103, is used for determining human face region based on the number of regional frame less than first threshold and greater than the features of skin colors of the cluster areas of Second Threshold.
The 3rd determining unit 106, be connected with described statistic unit 103, and the number that is used for the definite area frame is non-face zone less than the cluster areas of Second Threshold.
Second Threshold described in the present embodiment is the integer between [5~8], and described the second determining unit 105 comprises:
The 3rd detecting unit (not shown), be used for utilizing the Face Detection algorithm to detect the skin pixel point of the predeterminable area of the arbitrary regional frame of described cluster areas.
The 3rd determines the subelement (not shown), is used for number at described skin pixel point and during greater than the 5th threshold value, determines that described cluster areas is human face region than the value of the number of the pixel in described predeterminable area.
Described Face Detection algorithm is the Face Detection algorithm based on Gauss model or Bayesian model.Described the 5th threshold value is determined by test, the span of the 5th threshold value described in the present embodiment be [0.5~1), can get 0.5,0.6,0.7 etc. as, described the 5th threshold value.
In another embodiment, described the second determining unit comprises:
The first detecting unit, utilize the Face Detection algorithm to detect the skin pixel point of described cluster areas;
First determines subelement, be used for number at described skin pixel point than described cluster areas in the value of number of pixel during greater than the 3rd threshold value, determine that described cluster areas is human face region.
Described the 3rd threshold value is determined by test, and in the present embodiment, the span of described the 3rd threshold value is [0.3,0.8].Can get 0.3,0.4,0.5 etc. as, described the 3rd threshold value.
In another embodiment, described the second determining unit comprises:
The second detecting unit, utilize the Face Detection algorithm to detect the skin pixel point of arbitrary regional frame in described cluster areas;
Second determines subelement, be used for number at described skin pixel point than described arbitrary regional frame in the value of number of pixel during greater than the 4th threshold value, determine that described cluster areas is human face region.
Described the 4th threshold value is determined by test, in the present embodiment, the span of described the 4th threshold value be [0.5~1), can get 0.5,0.6,0.7,0.8 etc. as, described the 4th threshold value.
The course of work of the face of people described in the present embodiment pick-up unit, can carry out referring to above-mentioned method for detecting human face, no longer launches concrete detailed description the in detail herein.
In sum, technical scheme of the present invention has following beneficial effect at least:
A plurality of regional frames that obtain by people's face detection algorithm are carried out cluster analysis, add up the number of regional frame in each cluster areas, with this, detect human face region, for the method for detecting human face of skin color based, because it is subjected to illumination effect less, therefore loss is low; For the method for detecting human face based on Adaboost, owing to need not to gather a large amount of samples in the training process of detecting device, therefore reduced the complexity that people's face detects; And, after adopting people's face detection algorithm to obtain regional frame, first carry out cluster analysis, then the regional frame number in cluster areas is added up, also reduced the false drop rate that people's face detects.
In the number of the regional frame of cluster areas during less than first threshold and greater than Second Threshold, features of skin colors based on described cluster areas is determined human face region, compared to the method for detecting human face of skin color based with based on people's face detection algorithm of Adaboost, carry out people's face by the statistics of the regional frame number in conjunction with cluster areas and Face Detection algorithm and detect, further reduced loss and false drop rate that people's face detects.
While detecting skin pixel point, can detect the skin pixel point of described cluster areas, also can detect the skin pixel point of arbitrary regional frame in described cluster areas, can also detect the skin pixel point of the predeterminable area of arbitrary regional frame in described cluster areas, and then utilize the skin pixel point to determine human face region, therefore have very large dirigibility.
Although the present invention with preferred embodiment openly as above; but it is not to limit the present invention; any those skilled in the art without departing from the spirit and scope of the present invention; can utilize method and the technology contents of above-mentioned announcement to make possible change and modification to technical solution of the present invention; therefore; every content that does not break away from technical solution of the present invention;, to any simple modification, equivalent variations and modification that above embodiment does, all belong to the protection domain of technical solution of the present invention according to technical spirit of the present invention.