CN108319908A - A kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics - Google Patents
A kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics Download PDFInfo
- Publication number
- CN108319908A CN108319908A CN201810081168.8A CN201810081168A CN108319908A CN 108319908 A CN108319908 A CN 108319908A CN 201810081168 A CN201810081168 A CN 201810081168A CN 108319908 A CN108319908 A CN 108319908A
- Authority
- CN
- China
- Prior art keywords
- human face
- pixel
- face region
- image
- human
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/162—Detection; Localisation; Normalisation using pixel segmentation or colour matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics, including：Face datection is carried out to image to be detected using the trained human-face detector based on Pixel-level Differential Characteristics, obtains Preliminary detection human face region；If shooting environmental is night, Preliminary detection human face region is converted to Lab chrominance spaces, the colour cast factor is obtained using Lab chrominance spaces, when the colour cast factor is more than or equal to colour cast threshold value, Preliminary detection human face region is converted to YCrCb chrominance spaces and carries out colour of skin judgement, and then obtains new colour of skin threshold value；When the pixel number in Preliminary detection human face region in colour of skin threshold range is more than amount threshold, Preliminary detection human face region is human face region, is otherwise non-face region.The method for detecting human face of the present invention can obtain preferable Face datection effect under untethered environment, while detection speed has certain advantage.
Description
Technical field
The invention belongs to technical field of computer vision, more particularly, to a kind of based on the non-of Pixel-level Differential Characteristics
Constrained environment method for detecting human face.
Background technology
The target of Face datection is to find and orient the specific existence position of face in given image.Based on front
, the Face datection limited under background has been achieved for good effect, and has been widely used in every field.With need
That asks is continuously increased, and the face that only detection front limits under background has been unable to meet demand, when needs are in untethered environment
When accurately finding human face region in the photo of lower shooting, need the difficulty faced to have following several, for example, human face posture change
Change, the change of illumination condition is blocked, out of focus and low resolution etc..How precisely rapidly in the image of untethered environment shooting
The position of middle locating human face becomes the problem of numerous researchers pay close attention to.
The method for realizing Face datection can generally be summarized as first selecting suitable method to describe facial image offer
Information, then be compared with image to be detected by certain judge rule, to judge that detection zone belongs to face also right and wrong
Face.Existing Face datection algorithm includes mainly template matches, support vector machines, neural network and Adaboost, due to base
It is complex in the Face datection algorithm model of template matches and support vector machines, it is difficult for complicated scene training process,
Detection speed is therefore affected and detection result is unsatisfactory.Face datection algorithm training based on neural network needs a large amount of
Flag data, and the model that training obtains is complicated, detection speed cannot be satisfied requirement.
It can be seen that the prior art technical problem that there are detection speeds is slow, detection result is poor.
Invention content
For the disadvantages described above or Improvement requirement of the prior art, the present invention provides a kind of based on Pixel-level Differential Characteristics
Thus untethered environment method for detecting human face solves the prior art technical problem that there are detection speeds is slow, detection result is poor.
To achieve the above object, according to one aspect of the present invention, it provides a kind of based on the non-of Pixel-level Differential Characteristics
Constrained environment method for detecting human face, including：
(1) it treats detection image using human-face detector and carries out Face datection, obtain Preliminary detection human face region；
(2) judge whether shooting environmental is night by the brightness of image to be detected, it, will be preliminary if shooting environmental is night
Detection human face region is converted by RGB color degree space to Lab chrominance spaces, and the colour cast factor is obtained using Lab chrominance spaces, if shooting
Environment is not night, then enters step (4)；
(3) when the colour cast factor be more than or equal to colour cast threshold value when, by Preliminary detection human face region by RGB color degree space convert to
YCrCb chrominance spaces obtain new colour of skin threshold value according to the original colour of skin threshold value of YCrCb chrominance spaces and image to be detected；
(4) if shooting environmental, which is night and the colour cast factor, is more than or equal to colour cast threshold value, when in Preliminary detection human face region
When pixel number in new colour of skin threshold range is more than amount threshold, Preliminary detection human face region is human face region, otherwise
For non-face region, if shooting environmental, which is not night or the colour cast factor, is less than or equal to colour cast threshold value, when Preliminary detection face area
When pixel number in domain in original colour of skin threshold range is more than amount threshold, Preliminary detection human face region is face area
Otherwise domain is non-face region.
Human-face detector is the trained human-face detector based on Pixel-level Differential Characteristics, the instruction of the human-face detector
White silk includes：
Obtain the positive and negative samples that use of training, positive sample is facial image, negative sample be not comprising human face region it is non-by
Ambient image is limited, human-face detector is trained using positive and negative samples, extracts Pixel-level Differential Characteristics in training process, utilize pixel
Grade Differential Characteristics build the secondary tree of depth, are obtained depth secondary tree cascade using booststrap trained based on Pixel-level
The human-face detector of Differential Characteristics.
Further, training process also obtains difficult sample, using difficult sample as negative sample repetition training Face datection
Device accelerates training process.
Further, step (1) includes：
Detection image is treated using human-face detector and carries out Face datection, and the size of image to be detected is kept in detection process
It is constant, change the size of detection window, is based on minimum detection window amplification detection window, obtains the detection window of multiple sizes,
For the detection window of each size, using corresponding sliding step, Face datection is carried out in image to be detected, was being detected
Cheng Zhong calculates each detection window in every level-one score of human-face detector, adds up to every level-one score successively；If detection
Window is less than the segmentation threshold of current series in the cumulative score of every level-one of human-face detector, then the detection window is preliminary inspection
Survey human face region.
Further, the colour cast factor is：
Wherein, ε is the colour cast factor, and δ is the coloration mean value of Preliminary detection human face region, and ψ indicates Preliminary detection human face region
Coloration mean square deviation.
Further, the coloration mean value of Preliminary detection human face region is：
Wherein, δaIndicate the average value of a components in Lab chrominance spaces, δbB components is averaged in expression Lab chrominance spaces
Value.
Further, the coloration mean square deviation of Preliminary detection human face region is：
Wherein, ψaIndicate the coloration mean square deviation of a components in Lab chrominance spaces, ψbIndicate the color of b components in Lab chrominance spaces
Spend mean square deviation.
Further, the colour cast factor is for judging whether image to be detected occurs colour cast, as ε ＜ 1.5, image to be detected
Colour cast does not occur；As ε >=1.5, colour cast occurs for image to be detected, and the value of ε is bigger, the colour cast degree of image to be detected
It is more serious.
Further, δaWhen ＞ 0, illustrate that image to be detected is whole partially red, it is on the contrary then partially green；Work as δbWhen ＞ 0, illustrate to be checked
Altimetric image is whole partially yellow, otherwise partially blue.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect：
(1) present invention is combined as a result of by human-face detector based on Pixel-level Differential Characteristics with colour of skin judgement
Method has broken away from a fairly large number of defect of exclusive use Pixel-level Differential Characteristics human-face detector flase drop.First using based on pixel
The human-face detector of grade Differential Characteristics is detected, and Pixel-level Differential Characteristics is made full use of to have scale invariability, illumination variation
Robustness is blocked the characteristics of robustness, and Preliminary detection human face region is quickly obtained, and is reused the colour of skin and is judged according to colour cast situation tune
Whole threshold value carries out secondary judgement to the Preliminary detection human face region having detected that, excludes non-face region, reduce flase drop quantity.
Therefore, the method for the present invention has stronger adaptability for the variation of shooting environmental, while taking into account verification and measurement ratio and false drop rate.
(2) present invention is trained as a result of other congenic methods using difficult sample as the method for negative sample, have been broken away from
The defect that time is longer, training result convergence is slower.Using difficult negative sample as negative sample repetition training, by preceding primary training
Obtained difficult sample is preserved, as next time trained negative sample so that human-face detector can be in the stage earlier
Divide sample to handle difficulty, human-face detector is enable to restrain faster, accelerates training process.Therefore, the method for the present invention
The method training time that training time does not use difficult sample than other is shorter, and convergence rate faster, effectively increases face inspection
Survey the training process of device.
Description of the drawings
Fig. 1 is the stream of the untethered environment method for detecting human face provided in an embodiment of the present invention based on Pixel-level Differential Characteristics
Journey schematic diagram；
Fig. 2 is the training method flow signal provided in an embodiment of the present invention based on Pixel-level Differential Characteristics human-face detector
Figure；
Fig. 3 is that the secondary tree of depth provided in an embodiment of the present invention cascades flow diagram；
Fig. 4 is method for detecting human face flow diagram provided in an embodiment of the present invention；
Fig. 5 is the testing process schematic diagram provided in an embodiment of the present invention based on Pixel-level Differential Characteristics human-face detector；
Fig. 6 is that the colour of skin provided in an embodiment of the present invention judges flow diagram；
Fig. 7 (a) is the testing result for only using Pixel-level Differential Characteristics daytime provided in an embodiment of the present invention；
Fig. 7 (b) is daytime provided in an embodiment of the present invention Pixel-level Differential Characteristics and the colour of skin to be judged the detection knot combined
Fruit；
Fig. 7 (c) is the testing result for only using Pixel-level Differential Characteristics at night provided in an embodiment of the present invention；
Fig. 7 (d) is night provided in an embodiment of the present invention Pixel-level Differential Characteristics and the colour of skin to be judged the detection knot combined
Fruit；
Fig. 8 (a) is the testing result provided in an embodiment of the present invention that the first pictures are used with Seetaface；
Fig. 8 (b) is the testing result provided in an embodiment of the present invention that the second pictures are used with Seetaface；
Fig. 8 (c) is the testing result provided in an embodiment of the present invention that the first pictures are used with MTCNN；
Fig. 8 (d) is the testing result provided in an embodiment of the present invention that the second pictures are used with MTCNN；
Fig. 8 (e) is the testing result provided in an embodiment of the present invention that the first pictures are used with the method for the present invention；
Fig. 8 (f) is the testing result provided in an embodiment of the present invention that the second pictures are used with the method for the present invention.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
It does not constitute a conflict with each other and can be combined with each other.
As shown in Figure 1, a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics includes：
Step 101：The positive and negative samples that training uses are obtained, positive sample is facial image, includes not only positive face figure
Picture, further includes the side face of various angles, and wears ornament, makes various different expressions, male, women, various age levels
Secondary facial image intercepts obtained facial image under certain restriction, selectively carries out turning operation later,
Finally it is normalized to 24*24 sizes.Negative sample is the untethered ambient image not comprising human face region, including indoor and outdoor is various
Different scenes.Using random sliding step, random window size interception image in negative sample, 24*24 sizes are normalized to,
Overturning or transposition are selectively carried out, enriches negative sample collection so that the negative sample image of acquisition has more randomness.
Step 102：Human-face detector based on Pixel-level Differential Characteristics is trained using the training algorithm after optimization；It trained
Cheng Shouxian extracts Pixel-level Differential Characteristics, the secondary tree of depth is built using Pixel-level Differential Characteristics, finally by the secondary tree of depth
Cascade obtains the final trained human-face detector based on Pixel-level Differential Characteristics, due to being used in cascade process
Bootstrap modes, training process is very slow, and optimization method is to accelerate training process as negative sample using difficult sample；
Step 103：Face datection is carried out to image to be detected using trained human-face detector, obtains Preliminary detection people
Face region；Detection process is to keep image to be detected size constancy using the method for changing detection window size, changes detection window
The size of mouth.Selected minimum detection window, changes the size of window, until detection window and detection image according to a certain percentage
The minimum value of long width values is close.The window of each size is examined using certain sliding step in entire image
It surveys.In detection process, each detection window needs to calculate every level-one score in human-face detector successively, to every level-one
Score adds up, if under current series, the cumulative score of the window is less than the threshold value of current series, then directly sentences the window
It is set to the window not comprising human face region, subsequent calculating need not be carried out；Otherwise, until each grade of calculating of the human-face detector
It completes, threshold value is satisfied by requirement, this window is just determined as Preliminary detection human face region.
Step 104：Judge whether shooting environmental is night by the brightness of image to be detected, if shooting environmental is night,
Preliminary detection human face region is converted by RGB color degree space to Lab chrominance spaces, the colour cast factor is obtained using Lab chrominance spaces,
If shooting environmental is not night, 106 are entered step.
Step 105：When the colour cast factor is more than or equal to colour cast threshold value, by Preliminary detection human face region by RGB color degree space
To YCrCb chrominance spaces, colour of skin Clustering Effect on Cr, Cb component is preferable for conversion, when colour cast occurs, the cluster of Cr, Cb component
Overall offset is happened, new colour of skin threshold value is obtained according to the original colour of skin threshold value of YCrCb chrominance spaces and image to be detected；
Step 106：If shooting environmental is night and the colour cast factor is more than or equal to colour cast threshold value, when Preliminary detection human face region
In pixel in new colour of skin threshold range be colour of skin point, colour of skin point number is more than total pixel in Preliminary detection human face region
Points 30% when represent more than amount threshold, Preliminary detection human face region is human face region, is otherwise non-face region, if clapping
Take the photograph environment not and be night or the colour cast factor is less than or equal to colour cast threshold value, when in Preliminary detection human face region in original colour of skin threshold value
Pixel in range is colour of skin point, and colour of skin point number indicates when being more than 30% of total pixel number in Preliminary detection human face region
More than amount threshold, Preliminary detection human face region is human face region, is otherwise non-face region.
As shown in Fig. 2, step 102 further comprises following sub-step in the method for the present invention：
Step 201：Pixel-level Differential Characteristics are extracted, concrete operations are as follows：
In order to effectively improve detection speed in the extraction process of Pixel-level Differential Characteristics, we use and build before detection
The mode of vertical look-up table, all characteristic values being likely to occur are computed in advance, be normalized to [0,255], using pixel value as
Foundation is searched, is stored in two-dimensional array and is used as feature look-up table.
Step 202：The secondary tree of depth is built, concrete operations are as follows：
The building process of the secondary tree of depth uses Gentlle Adaboost algorithms, is updated to sample weights.Each
Node selects the Pixel-level Differential Characteristics and height segmentation threshold for error in classification minimum of sening as an envoy to.When selecting, an optimal pixel is differential
After dtex sign, one subseries is carried out to positive negative sample using current pixel grade Differential Characteristics and obtains all samples in current face
Score under detector, update sample weights formula are as follows：
Weight=exp (- y*Fx)
Wherein y indicates that the classification to sample, y=1 indicate that positive sample, y=-1 indicate that negative sample, Fx indicate obtaining for sample
Point.After carrying out weight update, the wrong weight for dividing sample is increased, the weight of correct classification samples is reduced, makes subsequent feature
It more concentrates on mistake and divides sample.
Step 203：The secondary tree of cascade deep, concrete operations are as follows：
Fig. 3 is that the secondary tree of depth cascades flow diagram, the mode of booststrap is used in cascade process, constantly
Excavate negative sample in ground.With the increase of training series, the performance of human-face detector is constantly promoted, and has better classification to sample
Ability, booststrap the time it takes can greatly increase, and seriously affect the speed of training.Meanwhile human-face detector is received
The speed held back is slower and slower, it is difficult to reach ideal detection result.It therefore, will using difficult sample as negative sample repetition training
The difficult sample that preceding primary training obtains is preserved, as next time trained negative sample so that human-face detector can be
Stage earlier divides sample to handle difficulty, and human-face detector is enable to restrain faster, accelerates training process.
As shown in figure 4, step 103 further comprises in the method for the present invention：
Using it is trained to the human-face detector based on Pixel-level Differential Characteristics be detected.Concrete operations are as follows：
It is illustrated in figure 4 the human-face detector testing process based on Pixel-level Differential Characteristics, detection window is big using changing
Small method keeps detection image size constancy, changes the size of detection window.Minimum detection window is training template size
24*24 changes the size of window according to a certain percentage, 1.2 times of ratio enlargement window is used in the present invention, until detection
Window and the minimum value of the long width values of detection image are close.For the window of each size, using certain sliding step, in whole picture
It is detected in image.For the size of sliding step, the processing carried out herein is as follows：It is sliding when window size is less than 40*40
Dynamic step-length is the 10% of window size；When window size is more than or equal to 40*40, sliding step is the 5% of window size.It is examining
During survey, each detection window needs to calculate every level-one score in human-face detector successively, to the score of every level-one into
Row is cumulative, if under current series, the cumulative score of the window is less than the threshold value of current series, then is directly determined as the window not
Include the window of human face region, subsequent calculating need not be carried out；Otherwise, it is completed until each grade of the human-face detector calculates,
Threshold value is satisfied by requirement, is just judged to this window to include the window of human face region.
Method due to using sliding window in the detection, testing result inevitably appear in same person on the face
There is the case where multiple detection blocks, therefore the human face region that detected overlapping is merged, selects closest to face
Region obtains Preliminary detection human face region.
Colour of skin judgement, removal flase drop region are carried out to having been detected by Preliminary detection human face region.Concrete operations are as follows：
It is illustrated in figure 5 the colour of skin and judges that flow first substantially judges shooting environmental, when shooting environmental is daytime
When, judge without colour cast and handle, area of skin color is judged according to original colour of skin threshold value；When shooting environmental is night
When, then it needs to carry out colour cast judgement, if judge that colour cast situation occurs in present image, selects specific threshold value to carry out the colour of skin and sentence
It is disconnected, if colour cast situation does not occur, original threshold value is selected to carry out colour of skin judgement.
Colour cast situation is carried out judging to need to convert Preliminary detection human face region to Lab chrominance spaces, L * component represents figure
The luminance information of picture, value range are [0,100], indicate ater to pure white in visual perception range；A components represent color
Information is spent, value range is [- 128,127], indicates green to red in visual perception range；B components are similarly represented as coloration
Information, value range are [- 128,127], indicate that blue arrives yellow in visual perception range.Due to Lab chrominance spaces with set
Standby selection is unrelated, so that it is applied in Color Image Retrieval relatively broad, can retain in the processing of image broad as possible
Colour gamut and abundant color.RGB color degree space, which is converted to Lab chrominance spaces, needs XYZ as intermediary, and transfer process is such as
Under.
As Y ＞ 0.008856, then have
As Y ＜ 0.008856, then have
Coloured image is converted from RGB color degree space to Lab chrominance spaces, the coloration mean value δ of image can be passed through
Carry out the colour cast degree of evaluation image with the ratio θ of coloration mean square deviation ψ.In order to keep reduced value more prominent, common pair can be taken to θ
Number obtains ε, and ε is usually called the colour cast factor, the colour cast degree of image, the bigger representative image of value of ε are described with the size of ε values
Colour cast it is more serious, otherwise illustrate image color cast lesser extent, or there is no colour cast situation, computational methods are as follows：
In above-mentioned formula, H, W indicate the length and width of image, δ respectivelya、δbTwo components of a, b in Lab chrominance spaces are indicated respectively
Average value, δ be image coloration mean value；ψa、ψbIndicate that the coloration mean square deviation of the two components, ψ indicate the coloration of image respectively
Mean square deviation, ε are the colour cast factors.Discriminatory analysis can be carried out to the colour cast situation of image, can select by statistics by these values
Determine discrimination threshold, as ε ＜ 1.5, colour cast does not occur for judgement present image；As ε >=1.5, color occurs for judgement present image
Partially, and if ε value it is bigger, the colour cast degree of image is more serious.Work as δaWhen ＞ 0, illustrate that present image is whole partially red, it is on the contrary then
It is partially green；Work as δbWhen ＞ 0, illustrate that present image is whole partially yellow, it is otherwise partially blue.
The colour of skin is carried out judging to use YCrCb chrominance spaces, transformation matrix as follows：
Cr, Cb component are clustered in different colour casts and are more concentrated, and thereby determine that judgment threshold, are carried out to colour of skin point
Judge.If shooting environmental, which is night and the colour cast factor, is more than or equal to colour cast threshold value, when in Preliminary detection human face region in new skin
When pixel number in color threshold range is more than amount threshold, Preliminary detection human face region is human face region, is otherwise inhuman
Face region, if shooting environmental, which is not night or the colour cast factor, is less than or equal to colour cast threshold value, when in Preliminary detection human face region
When pixel number in original colour of skin threshold range is more than amount threshold, Preliminary detection human face region is human face region, otherwise
For non-face region.
In order to assess the performance of the present invention, it is compared with several method for detecting human face.Pixel will be only used first
The method for detecting human face of grade Differential Characteristics judges that the method for detecting human face combined compares with by Pixel-level Differential Characteristics with the colour of skin
Compared with；Then the method for the present invention and existing method for detecting human face are compared, the method for participating in comparing has：By traditional characteristic and god
Method for detecting human face Seetaface, the deep learning method MTCNN being combined through network.
Fig. 7 (a) is the testing result for only using Pixel-level Differential Characteristics daytime；Fig. 7 (b) is that daytime is special by Pixel-level difference
Sign judges the testing result combined with the colour of skin；Fig. 7 (c) is the testing result for only using Pixel-level Differential Characteristics at night；Fig. 7 (d)
Pixel-level Differential Characteristics and the colour of skin are judged into the testing result combined for night；It can be seen that in the case of daytime and night originally
Inventive method can effectively reduce flase drop quantity.
Fig. 8 (a) is the testing result provided in an embodiment of the present invention that the first pictures are used with Seetaface；Fig. 8
(b) it is the testing result provided in an embodiment of the present invention that the second pictures are used with Seetaface；Fig. 8 (c) is that the present invention is real
The testing result that the first pictures are used with MTCNN of example offer is provided；Fig. 8 (d) is provided in an embodiment of the present invention for the
Two pictures use the testing result of MTCNN；Fig. 8 (e) uses this hair to be provided in an embodiment of the present invention for the first pictures
The testing result of bright method；Fig. 8 (f) is the detection provided in an embodiment of the present invention that the second pictures are used with the method for the present invention
As a result.As can be seen that there is preferable detection result for the larger face of fuzzy, angular deflection herein.
The present invention judges removal flase drop for the more problem of flase drop using the colour of skin.First using based on Pixel-level Differential Characteristics
Human-face detector be detected, using this feature have for illumination variation, block, obscure and the robust of low resolution
Property, image to be detected is detected, face candidate region is obtained, then is removed in candidate region by way of colour of skin judgement
Non-face region reaches the target for reducing flase drop, and colour of skin judgement will not be influenced by the different ethnic group colours of skin, different illumination conditions,
Judged simultaneously for the case where night colored light sources, adjusting thresholds are carried out according to colour cast situation.One kind proposed by the present invention
Untethered environment method for detecting human face based on Pixel-level Differential Characteristics makes full use of the advantage of Pixel-level Differential Characteristics, preferential to protect
Witness's face verification and measurement ratio, then removal flase drop region is judged by the colour of skin, false drop rate is reduced, while there is apparent speed when detecting
Advantage.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, all within the spirits and principles of the present invention made by all any modification, equivalent and improvement etc., should all include
Within protection scope of the present invention.
Claims (8)
1. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics, which is characterized in that including：
(1) it treats detection image using human-face detector and carries out Face datection, obtain Preliminary detection human face region；
(2) judge whether shooting environmental is night by the brightness of image to be detected, if shooting environmental is night, by Preliminary detection
Human face region is converted by RGB color degree space to Lab chrominance spaces, the colour cast factor is obtained using Lab chrominance spaces, if shooting environmental
It is not night, then enters step (4)；
(3) when the colour cast factor be more than or equal to colour cast threshold value when, by Preliminary detection human face region by RGB color degree space convert to
YCrCb chrominance spaces obtain new colour of skin threshold value according to the original colour of skin threshold value of YCrCb chrominance spaces and image to be detected；
(4) if shooting environmental, which is night and the colour cast factor, is more than or equal to colour cast threshold value, when in Preliminary detection human face region new
When pixel number in colour of skin threshold range is more than amount threshold, Preliminary detection human face region is human face region, is otherwise non-
Human face region, if shooting environmental, which is not night or the colour cast factor, is less than or equal to colour cast threshold value, when in Preliminary detection human face region
When pixel number in original colour of skin threshold range is more than amount threshold, Preliminary detection human face region is human face region, no
It is then non-face region；
The human-face detector is the trained human-face detector based on Pixel-level Differential Characteristics, the instruction of the human-face detector
White silk includes：
The positive and negative samples that training uses are obtained, positive sample is facial image, and negative sample is the untethered ring not comprising human face region
Border image trains human-face detector using positive and negative samples, Pixel-level Differential Characteristics is extracted in training process, differential using pixel
Point secondary tree of feature construction depth is obtained the secondary tree cascade of depth using booststrap trained based on Pixel-level difference
The human-face detector of feature.
2. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics as described in claim 1, feature
It is, the training process also obtains difficult sample, using difficult sample as negative sample repetition training human-face detector, accelerates instruction
Practice process.
3. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics as claimed in claim 1 or 2, special
Sign is that the step (1) includes：
Detection image is treated using human-face detector and carries out Face datection, and the size of image to be detected is kept not in detection process
Become, change the size of detection window, is based on minimum detection window amplification detection window, obtains the detection window of multiple sizes, it is right
In the detection window of each size, using corresponding sliding step, Face datection is carried out in image to be detected, in detection process
In, each detection window is calculated successively in every level-one score of human-face detector, is added up to every level-one score；If detecting window
Mouth is less than the segmentation threshold of current series in the cumulative score of every level-one of human-face detector, then the detection window is Preliminary detection
Human face region.
4. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics as claimed in claim 1 or 2, special
Sign is that the colour cast factor is：
Wherein, ε is the colour cast factor, and δ is the coloration mean value of Preliminary detection human face region, and ψ indicates the color of Preliminary detection human face region
Spend mean square deviation.
5. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics as claimed in claim 4, feature
It is, the coloration mean value of the Preliminary detection human face region is：
Wherein, δaIndicate the average value of a components in Lab chrominance spaces, δbIndicate the average value of b components in Lab chrominance spaces.
6. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics as claimed in claim 4, feature
It is, the coloration mean square deviation of the Preliminary detection human face region is：
Wherein, ψaIndicate the coloration mean square deviation of a components in Lab chrominance spaces, ψbIndicate that the coloration of b components in Lab chrominance spaces is equal
Variance.
7. a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics as claimed in claim 1 or 2, special
Sign is that the colour cast factor is for judging whether image to be detected occurs colour cast, and as ε ＜ 1.5, image to be detected is not sent out
Add lustre to partially；As ε >=1.5, colour cast occurs for image to be detected, and the value of ε is bigger, and the colour cast degree of image to be detected is tighter
Weight.
8. special such as a kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics described in claim 5 or 6
Sign is, the δaWhen ＞ 0, illustrate that image to be detected is whole partially red, it is on the contrary then partially green；Work as δbWhen ＞ 0, illustrate image to be detected
It is whole partially yellow, it is otherwise partially blue.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810081168.8A CN108319908A (en) | 2018-01-26 | 2018-01-26 | A kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810081168.8A CN108319908A (en) | 2018-01-26 | 2018-01-26 | A kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108319908A true CN108319908A (en) | 2018-07-24 |
Family
ID=62887279
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810081168.8A Pending CN108319908A (en) | 2018-01-26 | 2018-01-26 | A kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108319908A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165592A (en) * | 2018-08-16 | 2019-01-08 | 新智数字科技有限公司 | A kind of real-time rotatable method for detecting human face based on PICO algorithm |
| CN109389030A (en) * | 2018-08-23 | 2019-02-26 | 平安科技（深圳）有限公司 | Facial feature points detection method, apparatus, computer equipment and storage medium |
| CN110570627A (en) * | 2019-06-27 | 2019-12-13 | 河海大学 | Barrier disaster early warning device and monitoring early warning method based on captive balloon |
| CN111898470A (en) * | 2020-07-09 | 2020-11-06 | 武汉华星光电技术有限公司 | Device and method for extracting fingerprint outside screen and terminal |
| CN112082738A (en) * | 2020-08-24 | 2020-12-15 | 南京理工大学 | Performance evaluation test system and test method for color night vision camera |
| CN112669290A (en) * | 2020-12-30 | 2021-04-16 | 稿定（厦门）科技有限公司 | Image comparison method and device |
| US20210396869A1 (en) * | 2018-10-02 | 2021-12-23 | Nec Corporation | Ship detection system, method, and program |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS6462384A (en) * | 1987-09-03 | 1989-03-08 | Matsushita Electronics Corp | Fluorescent lamp of electric bulb color |
| CN106557750A (en) * | 2016-11-22 | 2017-04-05 | 重庆邮电大学 | It is a kind of based on the colour of skin and the method for detecting human face of depth y-bend characteristics tree |
-
2018
- 2018-01-26 CN CN201810081168.8A patent/CN108319908A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS6462384A (en) * | 1987-09-03 | 1989-03-08 | Matsushita Electronics Corp | Fluorescent lamp of electric bulb color |
| CN106557750A (en) * | 2016-11-22 | 2017-04-05 | 重庆邮电大学 | It is a kind of based on the colour of skin and the method for detecting human face of depth y-bend characteristics tree |
Non-Patent Citations (1)
| Title |
|---|
| 魏菁: "皮肤检测技术的研究与改进", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165592A (en) * | 2018-08-16 | 2019-01-08 | 新智数字科技有限公司 | A kind of real-time rotatable method for detecting human face based on PICO algorithm |
| CN109165592B (en) * | 2018-08-16 | 2021-07-27 | 新智数字科技有限公司 | Real-time rotatable face detection method based on PICO algorithm |
| CN109389030A (en) * | 2018-08-23 | 2019-02-26 | 平安科技（深圳）有限公司 | Facial feature points detection method, apparatus, computer equipment and storage medium |
| US20210396869A1 (en) * | 2018-10-02 | 2021-12-23 | Nec Corporation | Ship detection system, method, and program |
| CN110570627A (en) * | 2019-06-27 | 2019-12-13 | 河海大学 | Barrier disaster early warning device and monitoring early warning method based on captive balloon |
| CN111898470A (en) * | 2020-07-09 | 2020-11-06 | 武汉华星光电技术有限公司 | Device and method for extracting fingerprint outside screen and terminal |
| CN111898470B (en) * | 2020-07-09 | 2024-02-09 | 武汉华星光电技术有限公司 | Off-screen fingerprint extraction device and method and terminal |
| CN112082738A (en) * | 2020-08-24 | 2020-12-15 | 南京理工大学 | Performance evaluation test system and test method for color night vision camera |
| CN112669290A (en) * | 2020-12-30 | 2021-04-16 | 稿定（厦门）科技有限公司 | Image comparison method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108319908A (en) | A kind of untethered environment method for detecting human face based on Pixel-level Differential Characteristics | |
| CN105631880B (en) | Lane line dividing method and device | |
| CN100354875C (en) | Red eye moving method based on human face detection | |
| CN107491762B (en) | A kind of pedestrian detection method | |
| CN104834898B (en) | A kind of quality classification method of personage's photographs | |
| US8675960B2 (en) | Detecting skin tone in images | |
| CN105303152B (en) | A kind of human body recognition methods again | |
| CN109284669A (en) | Pedestrian detection method based on Mask RCNN | |
| CN109711281A (en) | A kind of pedestrian based on deep learning identifies again identifies fusion method with feature | |
| CN106934386B (en) | A kind of natural scene character detecting method and system based on from heuristic strategies | |
| CN105005766B (en) | A kind of body color recognition methods | |
| CN102194108B (en) | Smile face expression recognition method based on clustering linear discriminant analysis of feature selection | |
| CN104318266B (en) | A kind of image intelligent analyzes and processes method for early warning | |
| CN109657612A (en) | A kind of quality-ordered system and its application method based on facial image feature | |
| JP2000003452A (en) | Method for detecting face surface in digital picture, its detecting device, picture judging method, picture judging device and computer readable record medium | |
| CN102096823A (en) | Face detection method based on Gaussian model and minimum mean-square deviation | |
| CN109214336A (en) | A kind of vehicle window marker detection method and device | |
| CN109300110A (en) | A kind of forest fire image detecting method based on improvement color model | |
| CN109740572A (en) | A kind of human face in-vivo detection method based on partial color textural characteristics | |
| CN104143077B (en) | Pedestrian target search method and system based on image | |
| CN111461036A (en) | Real-time pedestrian detection method using background modeling enhanced data | |
| CN110222217A (en) | A kind of shoes watermark image search method based on sectionally weighting | |
| Hu et al. | Fast face detection based on skin color segmentation using single chrominance Cr | |
| CN117809053A (en) | DETR target detection method based on neighborhood relative difference | |
| CN115601358B (en) | Tongue picture image segmentation method under natural light environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180724 |
|
| RJ01 | Rejection of invention patent application after publication |