CN101013481A - Female body classification and identification method - Google Patents

Abstract

The invention involves a female body shape classification and identification method. Through factor analysis, it can obtain four local body factors objectively reflecting the human body feature: chest form factor, side form factor, buttocks form factor, abdomen form factor, according to the selected shape factor and its special levy parameters, and the local human body feature classification, in theory breakdown female body shape into the 54 categories, and combine with the existing clothing system to propose a new type of apparel tagging: No. /shape + partial-body factor, and identify the classification. Using the invention, clothing enterprises can fully understand the objectives of the body of the consumer group, can be quickly customized services to personalized consumers shape information, reasonable arrange facilitate apparel production, and reduce production costs. For consumers, the classification of figure enable consumers to buy clothing in clear selection criteria greatly reduces the time to test clothing, providing more on-line shopping convenience.

Description

A kind of female body classification and recognition methods

Technical field

The invention belongs to a kind of dress designing, manufacturing field, be specifically related to a kind of female body classification and recognition methods.

Background technology

Along with development of times, people's material life abundant, the clothes that meet the personal characteristics more and more are subjected to people's favor, so clothes MC (mass customized) arises at the historic moment, and become the new direction of apparel industry development gradually.Yet, since now the general human body sorting technique of rag trade and clothing identification exist classification thick, to characteristics such as the reflection of human body three-dimensional tridimensional form are more weak, demand that can't satisfying personalized clothing making, often need to rely on the pattern maker personal experience and judge that buman body type is to revise model, this is not only consuming time, consumption power, and deficient in stability, be difficult to reach the requirement of clothes MC industrialization rapid-action.Thereby the system of setting up women body build segmentation of a cover and identification automatically is very to be necessary.At present, the clothes researchist of the existing part of China recognizes that the scientific research of buman body type segmentation is worth, and has launched the research of related fields one after another.Representative has: Fang Fang (Donghua University) classifies the sagittal plane characteristic morphology of male body chest, tripe, abdomen, the back of the body, 5 parts of stern, each part is divided three classes separately, and finishes differentiation with 5 sagittal plane form factors of correspondence respectively as the physical characteristic parameter.This somatotype method has realized the segmentation to buman body type to a certain extent, yet because the characteristic parameter that is used to differentiate is two anthropological measuring point horizontal ranges, can only reflect the human figure on the one dimension interface.On the whole, the single characteristic parameter (degree of enclosing, length or its difference) of the many employings of more existing human body somatotype methods, must cause to reflect from multi-angle the 3 D stereo feature of human body, existing buman body type sorting technique rests on classification itself more simultaneously, do not combine with existing garment size sign, its result is difficult to directly apply in the garment production and realm of sale of present stage.

Summary of the invention

Technical matters to be solved by this invention provides a kind of suitable practical application, identifies the female body classification method that combines, reflects from multi-angle the 3 D stereo feature of human body with existing garment size.

Another technical matters to be solved by this invention provides a kind of based on above-mentioned female body classification, and the accuracy height is practical, is simple and easy to the women body build recognition methods of usefulness.

It is this female body classification method that the present invention solves the problems of the technologies described above the technical scheme that is adopted, and its classification step is:

1., by [TC ²] the following female chest data of 3-D measuring apparatus measurement: distance before the front distance, under bust girth, the chest measurement height, the under bust girth height, the front distance, the throat distance, the foreneck height, the chest measurement height, the side neck is put distance to chest, and chest is put to the waist joint, the colpus distance, preceding chest measurement, chest depth, chest measurement is grown crosswise and chest measurement,

2., utilize above-mentioned data based following formula to obtain describing the characteristic parameter K of chest ₁, K ₂, K ₃, K ₄, K ₅, K ₆:

K ₁=under bust girth angle of inclination=(distance before the distance-under bust girth of front)/(chest measurement height-under bust girth height)

K ₂=foreneck is to chest angle=(distance-throat, front distance)/(foreneck height-chest measurement height)

K ₃=side neck to chest point distance/chest is put to waist and is saved

K ₄=colpus distance/preceding chest measurement

K ₅=chest arrow transverse diameter ratio=chest depth/chest measurement is grown crosswise

K ₆=chest measurement

3., by [TC ²] 3-D measuring apparatus measures following women side data above the waist: rear neck point to waist saves, and FNP FRONT NECK POINT to waist saves, and collare is to the omoplate horizontal range, and collare is to the omoplate vertical range, distance behind the waist, distance after the across back, across back line height, the low back height, the under bust girth angle of inclination, foreneck is to the chest angle;

4., utilize above-mentioned step data based following formula 3. to obtain describing the women characteristic parameter S of side above the waist ₁, S ₂, S ₃, P ₁, P ₂:

S ₁=bent back of the body index=rear neck point to waist joint/FNP FRONT NECK POINT to waist saves

S ₂=omoplate be raised to collare angle=collare to omoplate horizontal range/collare to the omoplate vertical range

S ₃=omoplate is raised to low back angle=(behind the waist after the distance-across back apart from)/(across back line height-low back height)

P ₁=under bust girth angle of inclination

P ₂=foreneck is to the chest angle

5., by [TC ²] 3-D measuring apparatus measures the data of following female buttocks: stern is protruding, and waist is long, hip depth, stern is grown crosswise, hip circumference;

6., utilize above-mentioned step data based following formula 5. to obtain describing the characteristic parameter X of female buttocks ₁, X ₂, X ₃:

X ₁=rump plays rake angle=stern, and protruding/waist is long

X ₂=stern arrow transverse diameter ratio=hip depth/stern is grown crosswise

X ₃=hip circumference

7., by [TC ²] 3-D measuring apparatus measures the data of following women's belly: distance before the distance before the waist, abdomen, preceding waistline height, the abdominal circumference height, waistline, the abdomen circle, waist is long;

8., utilize above-mentioned step data based following formula 7. to obtain describing the characteristic parameter Y of female buttocks ₁, Y ₂, Y ₃, Y ₄:

Y ₁=abdomen salient angle=(the preceding distance of distance-abdomen before the waist)/(preceding waistline height-abdominal circumference height)

Y ₂=waistline

Y ₃=abdominal circumference

Y ₄=waist is long

9., the parameter at four positions of the comprehensive above-mentioned reflection women bodily form, by factorial analysis, obtain to objectively respond 4 local body-shape factorses of buman body type feature: chest form factor, side form factor, buttocks form factor, belly form factor, and each local build is divided into the 2-3 class, obtain following table:

Local body-shape factors	Factor sign		Class indication/implication
					Chest form factor side form factor buttocks form factor belly form factor	R C T F	The flat F1/ of the flat C1/ of R1/ bent back of the body T1/ is moderate	The moderate T2/ of the well-balanced C2/ of R2/ is well-balanced	The plentiful C3/ of the R3/ plentiful F2/ abdomen of T3/ of squaring one's shoulders is protruding

And then in conjunction with chest waist difference woman body is divided, therefore women's build classification is divided into: 3 * 3 * 3 * 2=54 kind, represent with En respectively, n=1～54 wherein, it is implication such as following table separately:

Classification	Chest	The side	Buttocks	Belly	Classification	Chest	The side	Buttocks	Belly
										E1	R1	C1	T1	F1	E28	R2	C2	T2	F2
E2	R1	C1	T1	F2	E29	R2	C2	T3	F1
										E3	R1	C1	T2	F1	E30	R2	C2	T3	F2
E4	R1	C1	T2	F2	E31	R2	C3	T1	F1
										E5	R1	C1	T3	F1	E32	R2	C3	T1	F2
E6	R1	C1	T3	F2	E33	R2	C3	T2	F1
										E7	R1	C2	T1	F1	E34	R2	C3	T2	F2
E8	R1	C2	T1	F2	E35	R2	C3	T3	F1
										E9	R1	C2	T2	F1	E36	R2	C3	T3	F2
E10	R1	C2	T2	F2	E37	R3	C1	T1	F1
										E11	R1	C2	T3	F1	E38	R3	C1	T1	F2
E12	R1	C2	T3	F2	E39	R3	C1	T2	F1
										E13	R1	C3	T1	F1	E40	R3	C1	T2	F2
E14	R1	C3	T1	F2	E41	R3	C1	T3	F1
										E15	R1	C3	T2	F1	E42	R3	C1	T3	F2
El6	R1	C3	T2	F2	E43	R3	C2	T1	F1
										E17	R1	C3	T3	F1	E44	R3	C2	T1	F2
E18	R1	C3	T3	F2	E45	R3	C2	T2	F1
										E19	R2	C1	T1	F1	E46	R3	C2	T2	F2
E20	R2	C1	T1	F2	E47	R3	C2	T3	F1
										E21	R2	C1	T2	F1	E48	R3	C2	T3	F2
E22	R2	C1	T2	F2	E49	R3	C3	T1	F1
										E23	R2	C1	T3	F1	E50	R3	C3	T1	F2
E24	R2	C1	T3	F2	E51	R3	C3	T2	F1
										E25	R2	C2	T1	F1	E52	R3	C3	T2	F2
E26	R2	C2	T1	F2	E53	R3	C3	T3	F1
										E27	R2	C2	T2	F1	E54	R3	C3	T3	F2

In conjunction with the definition of standard GB 1335-91, with the numerical value number of being of height, be type with the numerical value of chest measurement or waistline, add that above-mentioned build classification En represents the women body bodily form, has finished the classification of women body build.

It still is a kind of women body build recognition methods based on above-mentioned female body classification method that the present invention solves the problems of the technologies described above the technical scheme that is adopted, and adopts the recognition mode based on the BP neural network, and its identification step is:

1. with of the input of each local physical characteristic parameter, invited 10 clothes professional persons to projects as training network

The differentiation of giving a mark, with the output as training network of the evaluation result that obtains, the implication of the discrete numerical quantities of output is: chest is set " flat "=1, " well-balanced "=2, " plentiful "=3; Side " the bent back of the body "=1 above the waist, " moderate "=2, " squaring one's shoulders "=3; Buttocks is set " flat "=1, " well-balanced "=2, " plentiful "=3; Belly is set " moderate "=1, " abdomen is protruding "=2;

2., inputoutput data is carried out normalization respectively, standardized data is compressed between [1,1] according to the maximal value and the minimum value of inputoutput data,

The standardization formula:

${x_i}^{\′}=2*\frac{x_i-\mathrm{mix}\left\{x_i\right\}}{\max \left\{x_i\right\}-\min \left\{x_i\right\}}-1(i=\mathrm{1,2},...,262);$

3., all there are several samples at each position, with 70% as training sample, 30% as test sample book, the neuroid structure adopts a hidden layer, each parameters such as the node number of hidden layer and learning rate are as shown in the table, and three numerals in the network structure are represented the interstitial content of input layer, hidden layer and output layer successively respectively;

Toponym	Network structure	Parameter is provided with
						1r	epochs	goal	r
		Chest	6:10:3	0.05	100000					1e-2	20
The side	5:6:3					0.05	50000	1e-2	20
		Buttocks	3:6:3	0.05	50000					1e-2	20
Belly	4:5:1					0.05	50000	1e-2	20

4., above four sub neural networks are contacted, adopt several samples, still it is divided into two groups, 70% as training sample, and 30% as test sample book, we still adopt a hidden layer in experiment, the parameter of determining this neural network is 1r=0.05, epochs=100000, goal=1e-2, r=20, model structure is 16:10:10; Input parameter is: the under bust girth angle of inclination, foreneck is to the chest angle, side neck to chest point distance/chest is put to waist and is saved, colpus distance/preceding chest measurement, chest is vowed the transverse diameter ratio, chest measurement, bent back of the body index, omoplate is raised to the collare angle, omoplate is raised to the low back angle, rump plays rake angle, stern is vowed the transverse diameter ratio, hip circumference, the abdomen salient angle, waistline, abdominal circumference, waist is long, output parameter: the output of neuroid also is the output of four sub-networks in comprehensive front, obtain the vector of one 10 dimension like this, every three-dimensional representative is to the scoring at a position, differentiate for peaked node in the classification criterion at certain position three output nodes relevant with this position, the output at last comprehensive four positions is as net result, thus the nerve network system of foundation energy concentrated expression buman body type identification;

5., at last, validation test sample and neural network output valve compare.

The present invention compares with prior art and has the following advantages and effect: the characteristic parameter that 1) is used for each local somatotype and identification is a plurality of, derives from the different cross section form, can fully reflect the three-dimensional physical characteristic of each part.2) characteristic parameter is ratio and angle value substantially, focus on the similarity of investigating the buman body type mode of appearance, complement one another with number type of reflection build length, degree of enclosing size, be easy in conjunction with forming new number type mark---" number/type+local body-shape factors ", be garment production person and consumer's acceptance.3) the inherent learning functionality by the BP neural network has been simulated the process that the expert carries out build identification, makes the result of judge more objective, more reasonable, more stable.In conjunction with 3-D measuring apparatus, can realize the quick identification to individual build simultaneously, through facts have proved, accuracy rate is more than 85%.

Utilize this classification of woman's body and recognition methods, garment enterprise can comprehensively be understood the physical characteristic of target consumer group, need can obtain individual character consumer's build information of customize services again fast, so that rationally arrange the garment production type, reduces production costs.And concerning the consumer, the segmentation of build also makes the consumer when buying clothes clear selection criteria arranged, and can significantly reduce the time of fitting, and more shopping online is provided convenience.

Embodiment

The present invention is described in further detail below in conjunction with embodiment, and following examples are explanation of the invention and the present invention is not limited to following examples.

Embodiment: present embodiment is divided into female body classification method and two parts of recognition methods,

With the lower part female body classification method of the present invention has been described:

By factorial analysis, obtain to objectively respond 4 local body-shape factorses (chest form factor, side form factor, buttocks form factor, belly form factor) of buman body type feature, and each local build is divided into the 2-3 class.

Local body-shape factors	Factor sign		Class indication/implication
					Chest form factor side form factor buttocks form factor belly form factor	R C T F	The flat F1/ of the flat C1/ of R1/ bent back of the body T1/ is moderate	The moderate T2/ of the well-balanced C2/ of R2/ is well-balanced	The plentiful C3/ of the R3/ plentiful F2/ abdomen of T3/ of squaring one's shoulders is protruding

According to the principle that reflects the human body three-dimensional morphological feature comprehensively, according to the actual demand of clothing making, for each body-shape factors is set a plurality of characteristic parameters.These 4 groups totally 16 characteristic parameters mostly be angle value or ratio value greatly, and can pass through [TC ²] 3-D measuring apparatus directly or indirectly obtains.Specifically be provided with as follows:

The characteristic parameter of describing the chest position is: K={K ₁, K ₂, K ₃, K ₄, K ₅, K ₆}={ the under bust girth angle of inclination, foreneck is to the chest angle, side neck to chest point distance/chest is put to the waist joint, colpus distance/preceding chest measurement, chest arrow transverse diameter ratio, chest measurement }.

Wherein under bust girth angle of inclination and foreneck to chest angle has reflected the deflection of breast portion, and computing formula is:

Under bust girth angle of inclination=(distance before the distance-under bust girth of front)/(chest measurement height-under bust girth height);

Foreneck is to chest angle=(distance-throat, front distance)/(foreneck height-chest measurement height);

Side neck to chest point distance/chest is put the height and position that has reflected breast to waist joint, colpus distance/preceding chest measurement reflect two breast towards, chest arrow transverse diameter ratio=chest depth/chest measurement is grown crosswise, and has reflected the richness of chest.

The characteristic parameter of describing the upper half of human body side is: S={S ₁, S ₂, S ₃, P ₁, P ₂}={, song was carried on the back index, and omoplate is raised to the collare angle, and omoplate is raised to the low back angle, the under bust girth angle of inclination, foreneck is to the chest angle }.

Wherein bent back of the body index S ₁=rear neck point saves to waist to waist joint/FNP FRONT NECK POINT represents S ₁Big more, expression side physical characteristic is a lordosis, the bossed trend in back, and corresponding chest is more flat; Otherwise, S ₁More little, expression side physical characteristic be backbone more straight and upright, square one's shoulders, correspondingly the back is more flat.

Omoplate is raised to the collare angle, omoplate is raised to the degree of convexity that the low back angle reflects the back, and formula is respectively:

Omoplate be raised to collare angle=collare to omoplate horizontal range/collare to the omoplate vertical range;

Omoplate is raised to low back angle=(behind the waist after the distance-across back apart from)/(across back line height-low back height); Under bust girth angle of inclination and foreneck reflect the degree of squaring one's shoulders to the chest angle.

The characteristic parameter of describing buttocks is: X={X ₁, X ₂, X ₃}={, rump played rake angle, and stern is vowed transverse diameter ratio, hip circumference }.

Wherein rump plays rake angle=stern, and protruding/waist is long, is the parameter in shelves oblique line when design behind the trousers, and it can reflect the buttocks shape that warps well; Stern arrow transverse diameter ratio=hip depth/stern is grown crosswise and is represented the depth information of buttocks, for identical hip circumference, endures if stern arrow transverse diameter, shows buttocks than big, and pelvis is narrower; If stern vows that transverse diameter than little, shows that buttocks is more flat, the pelvis broad.

The parameter of describing the belly form is Y={Y ₁, Y ₂, Y ₃, Y ₄}={ the abdomen salient angle, waistline, abdominal circumference, waist is long.

Its midfield salient angle=(the preceding distance of distance-abdomen before the waist)/(preceding waistline height-abdominal circumference height).

In sum, the net result that women's physical characteristic parameter is selected is: under bust girth angle of inclination, foreneck to chest angle, side neck to chest point distance/chest are put and are carried on the back index, omoplate to waist joint, colpus distance/preceding chest measurement, chest arrow transverse diameter ratio, chest measurement, song and be raised to collare angle, omoplate and be raised to low back angle, rump to play rake angle, stern arrow transverse diameter ratio, hip circumference, abdomen salient angle, waistline, abdominal circumference, waist long, so just buman body type is described, fully reflected the multi-dimensional nature feature of buman body type from different orientation.Two individualities so just can think that both builds are similar as long as their these physical characteristics refer to that target value is similar.

According to selected body-shape factors and characteristic parameter thereof, and, theoretically women body build is subdivided into 54 classes, and in conjunction with existing garment size system to the classification of each body local physical characteristic, propose-kind of new garment size mark: " number/type+local body-shape factors ", specifically be defined as:

A, number/notion of type continues to use the definition of GB1335-81, i.e. the numerical value number of being of height is a type with the numerical value of chest measurement (or waistline);

B, local body-shape factors: women body build is classified at four positions such as chest, upper part of the body side, buttocks, bellies, and then woman body is divided in conjunction with chest waist difference, therefore the somatotype number is total in theory: 3 * 3 * 3 * 2=54 kind, represent with En respectively, n=1～54 wherein, it is implication such as following table separately:

Classification	Chest	The side	Buttocks	Belly	Classification	Chest	The side	Buttocks	Belly
										E1	R1	C1	T1	F1	E28	R2	C2	T2	F2
E2	R1	C1	T1	F2	E29	R2	C2	T3	F1
										E3	R1	C1	T2	F1	E30	R2	C2	T3	F2
E4	R1	C1	T2	F2	E31	R2	C3	T1	F1
										E5	R1	C1	T3	F1	E32	R2	C3	T1	F2
E6	R1	C1	T3	F2	E33	R2	C3	T2	F1
										E7	R1	C2	T1	F1	E34	R2	C3	T2	F2
E8	R1	C2	T1	F2	E35	R2	C3	T3	F1
										E9	R1	C2	T2	F1	E36	R2	C3	T3	F2
E10	R1	C2	T2	F2	E37	R3	C1	T1	F1
										E11	R1	C2	T3	F1	E38	R3	C1	T1	F2
E12	R1	C2	T3	F2	E39	R3	C1	T2	F1
										E13	R1	C3	T1	F1	E40	R3	C1	T2	F2
E14	R1	C3	T1	F2	E41	R3	C1	T3	F1
										E15	R1	C3	T2	F1	E42	R3	C1	T3	F2
E16	R1	C3	T2	F2	E43	R3	C2	T1	F1
										E17	R1	C3	T3	F1	E44	R3	C2	T1	F2
E18	R1	C3	T3	F2	E45	R3	C2	T2	F1
										E19	R2	C1	T1	F1	E46	R3	C2	T2	F2
E20	R2	C1	T1	F2	E47	R3	C2	T3	F1
										E21	R2	C1	T2	F1	E48	R3	C2	T3	F2
E22	R2	C1	T2	F2	E49	R3	C3	T1	F1
										E23	R2	C1	T3	F1	E50	R3	C3	T1	F2
E24	R2	C1	T3	F2	E51	R3	C3	T2	F1
										E25	R2	C2	T1	F1	E52	R3	C3	T2	F2
E26	R2	C2	T1	F2	E53	R3	C3	T3	F1
										E27	R2	C2	T2	F1	E54	R3	C3	T3	F2

With the lower part women body build of the present invention recognition methods has been described:

Because buman body type is to be made of composite factors such as people's body length, degree of enclosing, angles, thus actual to the identifying of buman body type be process to a fuzzy comprehensive evoluation of these factors, relying on manpower is to be difficult to finish real-time, objective, effective recognition.

The present invention's employing is based on the recognition mode of BP neural network, and concrete steps are as follows:

Make up sub neural network by being trained for each local build.

1), invited 10 clothes professional persons to projects differentiation of giving a mark, with the output of the evaluation result (as flat) that obtains as training network with of the input of each local physical characteristic parameter as training network; Because neural network can only be accepted numerical information, and can not accept language message, thereby linguistic variable is converted into discrete numerical quantities, as setting " flat "=1, " well-balanced "=2, " plentiful "=3, or setting " moderate "=1

2) in order to proofread and correct the input of different neurons to the influence of learning process and prevent the saturated of respective function, inputoutput data is carried out normalization respectively, standardized data is compressed between [1,1] according to the maximal value and the minimum value of inputoutput data.

The standardization formula:

${x_i}^{\′}=2*\frac{x_i-\mathrm{mix}\left\{x_i\right\}}{\max \left\{x_i\right\}-\min \left\{x_i\right\}}-1(i=\mathrm{1,2},...,262)$

The evaluation that provides according to the expert in the training for position build identifications such as chest, side, buttocks obtains three classes, is respectively " flat ", " moderate ", " plentiful ", and each input parameter is mapped to them between [1,1] by regular method.The output layer of this neuroid adopts three nodes, and its target output is respectively [+1 ,-1 ,-1] for three classifications, and [1, + 1 ,-1], [1 ,-1, + 1], for the belly position, the evaluation result that provides according to the expert is divided into two classes, is respectively " moderate ", " abdomen is protruding ".Each input parameter is mapped to them between [1,1] by regular method.The output layer of this neuroid adopts a node, for two classifications its target output be respectively+1 ,-1}.To export among the result, maximal value is exported the classification of pairing classification as classification.

According to foregoing definite method to network model, all there are 262 samples at each position, still with 70% as training sample, has 183 samples; 30% as test sample book, has 79 samples.The neuroid structure adopts a hidden layer, and each parameters such as the node number of hidden layer and learning rate are as shown in the table, and three numerals in the network structure are represented the interstitial content of input layer, hidden layer and output layer respectively.

Toponym	Network structure	Parameter is provided with
						1r	epochs	goal	r
		Chest	6:10:3	0.05	100000					1e-2	20
The side	5:6:3					0.05	50000	1e-2	20
		Buttocks	3:6:3	0.05	50000					1e-2	20
Belly	4:5:1					0.05	50000	1e-2	20

Through checking, the training sample at four positions and the accuracy rate of test sample book are all more than 85%, and wherein the identification to chest, buttocks and belly all reaches more than 90%, has the better prediction effect.

The comprehensive identification of structure neural network

Above four sub neural networks are contacted, set up the nerve network system of energy concentrated expression buman body type identification.Adopt 262 samples, still it is divided into two groups, 70% as training sample, has 183 samples; 30% as test sample book, has 79 samples.We still adopt a hidden layer in experiment.

(1) input parameter: under bust girth angle of inclination, foreneck to chest angle, side neck to chest point distance/chest are put to waist joint, colpus distance/preceding chest measurement, chest and are vowed that transverse diameter ratio, chest measurement, bent back of the body index, omoplate are raised to collare angle, omoplate and are raised to low back angle, rump to play rake angle, stern arrow transverse diameter ratio, hip circumference, abdomen salient angle, waistline, abdominal circumference, waist long.

(2) output parameter: the output of neuroid also is the output of four sub-networks in comprehensive front, obtains the vector of one 10 dimension like this, and every three-dimensional representative is to the scoring at a position, for example, someone breast shape is well-balanced, and the side form is moderate, the buttocks form is plentiful, and the belly form is moderate, and then its target output vector is [1,1 ,-1 ,-1,1 ,-1 ,-1,-1,1 ,-1]., to differentiate for peaked node in the classification criterion at certain position three output nodes relevant with this position, the output at last comprehensive four positions is as net result.

(3) validation test sample and neural network output valve compare.Reaching optimum with error criterion is purpose, and the parameter of determining this neural network is 1r=0.05, epochs=100000, and goal=1e-2, r=20, model structure is 16:10:10.

Application example:

Individual first by 3D anthropometric scanning after, computing machine will generate needed 16 the human body physical characteristic parameter values of identification fast, again these numerical value directly are directed in the neural network model of making in advance, just can obtain corresponding evaluation result (being output vector), for example output vector is [1,1 ,-1,-1,1 ,-1 ,-1,1,1 ,-1] time, according to maximum subjection principle, the local body-shape factors R2C2T3F1 that obtains this individuality is E29, represents this consumer to belong to the build that the chest moulding is well-balanced, side pose is moderate, the buttocks moulding is plentiful, belly is moderate.As this individuality former type is 170/92B, then obtains its new garment size and is labeled as: " 170/92E29 ".

Claims (2)

1, a kind of female body classification method, it is characterized in that: classification step is:

1., by [TC ²] the following female chest data of 3-D measuring apparatus measurement: distance before the front distance, under bust girth, the chest measurement height, the under bust girth height, the front distance, the throat distance, the foreneck height, the chest measurement height, the side neck is put distance to chest, and chest is put to the waist joint, the colpus distance, preceding chest measurement, chest depth, chest measurement is grown crosswise and chest measurement,

2., utilize above-mentioned data based following formula to obtain describing the characteristic parameter K of chest ₁, K ₂, K ₃, K ₄, K ₅, K ₆:

K ₁=under bust girth angle of inclination=(distance before the distance-under bust girth of front)/(chest measurement height-under bust girth height)

K ₂=foreneck is to chest angle=(distance-throat, front distance)/(foreneck height-chest measurement height)

K ₃=side neck to chest point distance/chest is put to waist and is saved

K ₄=colpus distance/preceding chest measurement

K ₅=chest arrow transverse diameter ratio=chest depth/chest measurement is grown crosswise

K ₆=chest measurement

3., by [TC ²] 3-D measuring apparatus measures following women side data above the waist: rear neck point to waist saves, and FNP FRONT NECK POINT to waist saves, and collare is to the omoplate horizontal range, and collare is to the omoplate vertical range, distance behind the waist, distance after the across back, across back line height, the low back height, the under bust girth angle of inclination, foreneck is to the chest angle;

4., utilize above-mentioned step data based following formula 3. to obtain describing the women characteristic parameter S of side above the waist ₁, S ₂, S ₃, P ₁, P ₂:

S ₁=bent back of the body index=rear neck point to waist joint/FNP FRONT NECK POINT to waist saves

S ₂=omoplate be raised to collare angle=collare to omoplate horizontal range/collare to the omoplate vertical range

S ₃=omoplate is raised to low back angle=(behind the waist after the distance-across back apart from)/(across back line height-low back height)

P ₁=under bust girth angle of inclination

P ₂=foreneck is to the chest angle

5., by [TC ²] 3-D measuring apparatus measures the data of following female buttocks: stern is protruding, and waist is long, hip depth, stern is grown crosswise, hip circumference;

6., utilize above-mentioned step data based following formula 5. to obtain describing the characteristic parameter X of female buttocks ₁, X ₂, X ₃:

X ₁=rump plays rake angle=stern, and protruding/waist is long

X ₂=stern arrow transverse diameter ratio=hip depth/stern is grown crosswise

X ₃=hip circumference

7., by [TC ²] 3-D measuring apparatus measures the data of following women's belly: distance before the distance before the waist, abdomen, preceding waistline height, the abdominal circumference height, waistline, abdominal circumference, waist is long;

8., utilize above-mentioned step data based following formula 7. to obtain describing the characteristic parameter Y of female buttocks ₁, Y ₂, Y ₃, Y ₄:

Y ₁=abdomen salient angle=(the preceding distance of distance-abdomen before the waist)/(preceding waistline height-abdominal circumference height)

Y ₂=waistline

Y ₃=abdominal circumference

Y ₄=waist is long

9., the parameter at four positions of the comprehensive above-mentioned reflection women bodily form, by factorial analysis, obtain to objectively respond 4 local body-shape factorses of buman body type feature: chest form factor, side form factor, buttocks form factor, belly form factor, and each local build is divided into the 2-3 class, obtain following table:

Local body-shape factors Factor sign Class indication/implication Chest form factor R R1/ is flat R2/ is well-balanced R3/ is plentiful The side form factor C The bent back of the body of C1/ C2/ is moderate C3/ squares one's shoulders The buttocks form factor T T1/ is flat T2/ is well-balanced T3/ is plentiful The belly form factor F F1/ is moderate The F2/ abdomen is protruding

And then in conjunction with chest waist difference woman body is divided, therefore women's build classification is divided into: 3 * 3 * 3 * 2=54 kind, represent with En respectively, n=1～54 wherein, it is implication such as following table separately:

Classification Chest The side Buttocks Belly Classification Chest The side Buttocks Belly E1 R1 C1 T1 F1 E28 R2 C2 T2 F2 E2 R1 C1 T1 F2 E29 R2 C2 T3 F1 E3 R1 C1 T2 F1 E30 R2 C2 T3 F2 E4 R1 C1 T2 F2 E31 R2 C3 T1 F1 E5 R1 C1 T3 F1 E32 R2 C3 T1 F2 E6 R1 C1 T3 F2 E33 R2 C3 T2 F1 E7 R1 C2 T1 F1 E34 R2 C3 T2 F2 E8 R1 C2 T1 F2 E35 R2 C3 T3 F1 E9 R1 C2 T2 F1 E36 R2 C3 T3 F2 E10 R1 C2 T2 F2 E37 R3 C1 T1 F1 E11 R1 C2 T3 F1 E38 R3 C1 T1 F2 E12 R1 C2 T3 F2 E39 R3 C1 T2 F1 E13 R1 C3 T1 F1 E40 R3 C1 T2 F2 E14 R1 C3 T1 F2 E41 R3 C1 T3 F1 E15 R1 C3 T2 F1 E42 R3 C1 T3 F2 E16 R1 C3 T2 F2 E43 R3 C2 T1 F1 E17 R1 C3 T3 F1 E44 R3 C2 T1 F2 E18 R1 C3 T3 F2 E45 R3 C2 T2 F1 E19 R2 C1 T1 F1 E46 R3 C2 T2 F2 E20 R2 C1 T1 F2 E47 R3 C2 T3 F1 E21 R2 C1 T2 F1 E48 R3 C2 T3 F2 E22 R2 C1 T2 F2 E49 R3 C3 T1 F1 E23 R2 C1 T3 F1 E50 R3 C3 T1 F2

E24 R2 C1 T3 F2 E51 R3 C3 T2 F1 E25 R2 C2 T1 F1 E52 R3 C3 T2 F2 E26 R2 C2 T1 F2 E53 R3 C3 T3 F1 E27 R2 C2 T2 F1 E54 R3 C3 T3 F2

In conjunction with the definition of standard GB 1335-91, with the numerical value number of being of height, be type with the numerical value of chest measurement or waistline, add that above-mentioned build classification En represents the women body bodily form, has finished the classification of women body build.

2, the women body build recognition methods of female body classification method according to claim 1 is characterized in that: adopt the recognition mode based on the BP neural network, its identification step is:

1. with of the input of each local physical characteristic parameter as training network, invited 10 clothes professional persons to projects differentiation of giving a mark, with the output as training network of the evaluation result that obtains, the implication of the discrete numerical quantities of output is:

Chest is set " flat "=1, " well-balanced "=2, " plentiful "=3; Side " the bent back of the body "=1 above the waist, " moderate "=2, " squaring one's shoulders "=3; Buttocks is set " flat "=1, " well-balanced "=2, " plentiful "=3; Belly is set " moderate "=1, " abdomen is protruding "=2;

2., inputoutput data is carried out normalization respectively, standardized data is compressed between [1,1] according to the maximal value and the minimum value of inputoutput data,

The standardization formula:

$x_i^{\′}=2*\frac{x_i-\min \left\{x_i\right\}}{\max \left\{x_i\right\}-\min \left\{x_i\right\}}-1,(i=\mathrm{1,2},...,262);$

3., all there are several samples at each position, with 70% as training sample, 30% as test sample book, the neuroid structure adopts a hidden layer, each parameters such as the node number of hidden layer and learning rate are as shown in the table, and three numerals in the network structure are represented the interstitial content of input layer, hidden layer and output layer successively respectively;

Toponym Network structure Parameter is provided with lr epochs goal r Chest 6:10:3 0.05 100000 le-2 20 The side 5:6:3 0.05 50000 le-2 20 Buttocks 3:6:3 0.05 50000 le-2 20 Belly 4:5:1 0.05 50000 le-2 20

4., above four sub neural networks are contacted, adopt several samples, still it is divided into two groups, 70% as training sample, and 30% as test sample book, we still adopt a hidden layer in experiment, the parameter of determining this neural network is lr=0.05, epochs=100000, goal=le-2, r=20, model structure is 16:10:10; Input parameter is: the under bust girth angle of inclination, foreneck is to the chest angle, side neck to chest point distance/chest is put to waist and is saved, colpus distance/preceding chest measurement, chest is vowed the transverse diameter ratio, chest measurement, bent back of the body index, omoplate is raised to the collare angle, omoplate is raised to the low back angle, rump plays rake angle, stern is vowed the transverse diameter ratio, hip circumference, the abdomen salient angle, waistline, abdominal circumference, waist is long, output parameter: the output of neuroid also is the output of four sub-networks in comprehensive front, obtain the vector of one 10 dimension like this, every three-dimensional representative is to the scoring at a position, differentiate for peaked node in the classification criterion at certain position three output nodes relevant with this position, the output at last comprehensive four positions is as net result, thus the nerve network system of foundation energy concentrated expression buman body type identification;

5., at last, validation test sample and neural network output valve compare.