CN103440646B - Similarity acquisition methods for distribution of color and grain distribution image retrieval - Google Patents

Abstract

The invention discloses the similarity acquisition methods for distribution of color and grain distribution image retrieval, relate to image retrieval technologies.Technical key point includes: extract Color Distribution Features and the grain distribution feature of input picture；Calculate respectively the Color Distribution Features of described input picture with data base the similarity of the Color Distribution Features of every piece image, obtain Color Distribution Features similarity Sa(i between every piece image in input picture and data base)；Calculate respectively the grain distribution feature of described input picture with data base the similarity of the grain distribution feature of every piece image, obtain the grain distribution characteristic similarity Sb(i between every piece image in input picture and data base)；Utilize formula S (i)=Wa × Sa(i)+Wb × Sb(i), combination similarity S(i of every piece image in calculating input image and data base).

Description

Similarity acquisition methods for distribution of color and grain distribution image retrieval

Technical field

The present invention relates to image retrieval technologies, especially a kind of for distribution of color and grain distribution The similarity acquisition methods of image retrieval.

Background technology

In recent years, along with developing rapidly of mobile Internet, application of taking pictures obtains the biggest sending out Exhibition space, the acquisition of photo and storage become the easiest.Increasing along with picture data explosion type Long, user is in the urgent need to the retrieval of photo and the automatic technology of arrangement.Existing image is examined Rope technology will rely on the training sample of prestored digital image in data base and try to achieve similarity.And current cloud The photo of storage is substantially from the photo of the various scenes of various user shooting, not The training sample that retrievable display is labelled with.Thus the inconvenience of existing image retrieval technologies is directly It is applied in the retrieval of cloud storage image.

Summary of the invention

The technical problem to be solved is: for the problem of above-mentioned existence, it is provided that a kind of It is applicable to the similarity acquisition methods of cloud storage distribution of color and grain distribution image retrieval.

The similarity acquisition side for distribution of color and grain distribution image retrieval that the present invention provides Method, it is characterised in that including:

Step 1: extract Color Distribution Features and the grain distribution feature of input picture；

Step 2: the Color Distribution Features calculating described input picture respectively is each with data base The similarity of the Color Distribution Features of width image, obtains input picture and each width figure in data base Color Distribution Features similarity Sa(i between Xiang), i takes 0,1,2 ... database images is total Number-1；

Calculate respectively the grain distribution feature of described input picture with data base every piece image The similarity of grain distribution feature, obtains in input picture and data base between every piece image Grain distribution characteristic similarity Sb(i), i takes 0,1,2 ... database images sum-1；

Step 3: utilize formula S (i)=Wa × Sa(i)+Wb × Sb(i), i takes 0,1,2 ... Database images sum-1, Wa, Wb are weight coefficient and Wa+Wb=1, calculate input figure Combination similarity S(i as piece image every with data base).

Preferably, the acquisition methods of described Color Distribution Features includes:

Step 201: image is transformed into hsv color space, obtains image I；

Step 202: the H of each pixel of image, S, V component are mapped as color feature value G: G=Q_s*Q_v*H+Q_v*S+V；The span of three passages in hsv color space is carried out Interval division, is respectively divided into H_i,S_j,V_k, wherein 0≤i≤Q_h,0≤j≤Q_s,0≤k≤Q_v, Q_h,Q_s,Q_vRepresent the divided interval sum of three passages in hsv color space respectively；

Step 203: the eigenvalue distribution situation of each pixel in statistical picture: travel through each The color feature value of pixel, statistics falls into the pixel number of each distribution of color histogram Amount, is respectively divided by image pixel by the pixel quantity falling into each distribution of color histogram Point sum, obtains normalized Color Distribution Features hist (x), and wherein x represents distribution of color Nogata Figure interval.

Preferably, the acquisition methods of described Color Distribution Features also includes:

Divide an image into N block；In described step 203: each pixel in statistical picture Eigenvalue distribution situation: traveling through the eigenvalue of each pixel, statistics falls into each color and divides The pixel quantity of cloth histogram, and will not be the pixel statistics two in image boundary block Secondary；The pixel quantity falling into each distribution of color histogram is respectively divided by image pixel Point sum, obtains normalized Color Distribution Features hist (x), and wherein x represents color histogram.

Preferably, the acquisition methods of described grain distribution feature includes:

Step 301: convert the image into gray-scale map, obtains image L；

Step 302: with the template of a size of 3 pixel × 3 pixels, travels through described image L, To the LBP feature of each template, the method wherein obtaining template LBP feature includes:

The gray value of 9 pixels in note template is p_i(0≤i≤8), the wherein picture of template center Element gray value is designated as p₀；The gray value of other pixel in template is deducted p₀Obtain:

g_i=p_i-p₀,(1≤i≤8)；

To each calculated g_iCarry out binary conversion treatment: if g_i>=0 makes g_i=1, otherwise g_i=0；Will be located in the g of the pixel of position i_iValue expands to 82 system numbers, obtains LBP (i) special Levy as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: obtain the LBP of the invariable rotary of each template^riFeature；Wherein obtain template The LBP of invariable rotary^riThe method of feature includes:

To each LBP (i) of template by carrying out shifting function, 8 binary numbers can be obtained respectively According to, take a wherein minimum LBP as invariable rotary^ri(i) feature:

1≤i≤8 in formula, ROR represents shifting function, q Represent shift amount；

Step 304: add up the LBP of each invariable rotary in each template^riThe distribution feelings of (i) feature Condition: travel through the LBP of each invariable rotary of each template^riI () eigenvalue, statistics falls into each stricture of vagina The pixel quantity that reason distribution histogram is interval, then each grain distribution histogram will be fallen into Pixel quantity be respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature Hist (y), wherein y represents grain distribution histogram.

Preferably, the method calculating Color Distribution Features similarity Sa in described step 2 includes:

Step 401: utilize formulaCalculate Color Distribution Features Similarity, wherein hist₁X () is the Color Distribution Features of piece image, hist₂X () is the second width The Color Distribution Features of image.

Preferably, the method calculating grain distribution characteristic similarity Sb in described step 2 includes:

Step 401: utilize formulaCalculate grain distribution feature phase Like degree, wherein hist₁Y () is the grain distribution feature of piece image, hist₂Y () is the second width figure The grain distribution feature of picture.

Preferably, described Wa > Wb.

The present invention also protects a kind of similarity acquisition side for grain distribution image search method Method, including:

Step 1: extract the grain distribution feature of input picture；

Step 2: the grain distribution feature calculating described input picture respectively is each with data base The similarity of the grain distribution feature of width image, obtains input picture and each width figure in data base Grain distribution characteristic similarity Sb(i between Xiang), i takes 0,1,2 ... database images is total Number-1；

The acquisition methods of described grain distribution feature includes:

Step 301: convert the image into gray-scale map, obtains image L；

Step 302: with the template of a size of 3 pixel × 3 pixels, travels through described image L, To the LBP feature of each template, the method wherein obtaining template LBP feature includes:

The gray value of 9 pixels in note template is p_i(0≤i≤8), the wherein picture of template center Element gray value is designated as p₀；The gray value of other pixel in template is deducted p₀Obtain:

g_i=p_i-p₀,(1≤i≤8)；

To each calculated g_iCarry out binary conversion treatment: if g_i>=0 makes g_i=1, otherwise g_i=0；Will be located in the g of the pixel of position i_iValue expands to 82 system numbers, obtains LBP (i) special Levy as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

Step 303: obtain the LBP of the invariable rotary of each template^riFeature；Wherein obtain template The LBP of invariable rotary^riThe method of feature includes:

To each LBP (i) of template by carrying out shifting function, 8 binary numbers can be obtained respectively According to, take a wherein minimum LBP as invariable rotary^ri(i) feature:

1≤i≤8 in formula, ROR represents shifting function, q Represent shift amount；

Step 304: add up the LBP of each invariable rotary in each template^riThe distribution feelings of (i) feature Condition: travel through the LBP of each invariable rotary of each template^riI () eigenvalue, statistics falls into each stricture of vagina The pixel quantity that reason distribution histogram is interval, then each grain distribution histogram will be fallen into Pixel quantity be respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature Hist (y), wherein y represents grain distribution histogram.

The present invention also protect above-mentioned in the acquisition methods of grain distribution feature.

In sum, owing to have employed technique scheme, the invention has the beneficial effects as follows:

The image similarity acquisition methods that the present invention relates to is not required to carry out image any it is assumed that also Need not substantial amounts of mark sample training model, there is easily realization, calculate fireballing advantage.

Accompanying drawing explanation

Examples of the present invention will be described by way of reference to the accompanying drawings, wherein:

Fig. 1 is that in the present invention, Color Distribution Features extracts flow chart.

Fig. 2 is grain distribution feature extraction flow chart in the present invention.

Fig. 3 is image retrieval flow chart in the present invention.

Detailed description of the invention

All features disclosed in this specification, or disclosed all methods or during step, In addition to mutually exclusive feature and/or step, all can combine by any way.

Any feature disclosed in this specification, unless specifically stated otherwise, all can by other equivalence or The alternative features with similar purpose is replaced.I.e., unless specifically stated otherwise, each feature is only It it is an example in a series of equivalence or similar characteristics.

The invention provides a kind of similarity for distribution of color and grain distribution image retrieval to obtain Access method, its concrete steps include:

Step 1: extract Color Distribution Features and the grain distribution feature of input picture；

Step 2: the Color Distribution Features calculating described input picture respectively is each with data base The similarity of the Color Distribution Features of width image, obtains input picture and each width figure in data base Color Distribution Features similarity Sa(i between Xiang), i takes 0,1,2 ... database images is total Number-1；

Calculate respectively the grain distribution feature of described input picture with data base every piece image The similarity of grain distribution feature, obtains in input picture and data base between every piece image Grain distribution characteristic similarity Sb(i), i takes 0,1,2 ... database images sum-1；

Step 3: utilize formula S (i)=Wa × Sa(i)+Wb × Sb(i), i takes 0,1, 2... database images sum-1, Wa, Wb are weight coefficient and Wa+Wb=1, calculate input Image is combination similarity S(i of every piece image with data base).Owing to people are in general feelings More concerned with color under condition, therefore as one preferred embodiment, weight coefficient Wa > Wb.

Such as Fig. 3, when the combination obtained between input picture piece image every to data base is similar After degree, being ranked up each similarity, similarity the biggest explanation two width image is the most similar, I Can rule of thumb set a threshold value, will be greater than combining similarity more than all numbers of this threshold value Export according to the image in storehouse, as retrieval result.

Such as Fig. 1, in an embodiment of the invention, the acquisition methods of Color Distribution Features includes:

Step 201: image is transformed into hsv color space, obtains image I；In general scheme Sheet is RGB color, the picture of RGB color is transformed into hsv color space and is Technology well known in the art, does not repeats them here its detailed process.

Step 202: by the H of each pixel of image, S, V component according to formula G=Q_s*Q_v*H+Q_v* S+V mapping relations are converted to color feature value G；Wherein, Q_h,Q_s,Q_v Definition be such that carrying out the span of three passages in hsv color space interval drawing Point, it is respectively divided into H_i,S_j,V_k, wherein 0≤i≤Q_h,0≤j≤Q_s, 0 ￡ k≤Q_v, Q_h,Q_s,Q_vPoint Biao Shi the divided interval sum of three passages in hsv color space；

Step 203: the eigenvalue distribution situation of each pixel in statistical picture: travel through each The color feature value of pixel, statistics falls into the pixel number of each distribution of color histogram Amount, is respectively divided by image pixel by the pixel quantity falling into each distribution of color histogram Point sum, obtains normalized Color Distribution Features hist (x), and wherein x represents distribution of color Nogata Figure interval.

Those skilled in the art all know, and distribution of color rectangular histogram is by the color characteristic of entire image It is divided into some intervals, then describes different color by each pixel in the situation of each interval distribution Ratio shared in entire image.

In view of the implication expressed by piece image, it is often positioned in the region of near image boundaries not The most important, we are more concerned with scheming the content that non-borderline region is expressed.Therefore, the present invention another In individual embodiment, the acquisition methods of described Color Distribution Features also includes:

Divide an image into N block, such as N equal to 36；It is included in the picture in image boundary block Element is only added up once, and rest of pixels is added up twice.Specifically, in described step 203 In: the eigenvalue distribution situation of each pixel in statistical picture: travel through the spy of each pixel Value indicative, statistics falls into the pixel quantity of each distribution of color histogram；And will not be figure As the pixel in boundary block is added up twice, i.e. be not the pixel in image boundary block when having Eigenvalue when falling into a certain distribution of color histogram, the pixel number in this interval will be fallen into Amount adds 2；When having, to be that the eigenvalue of pixel in image boundary block falls into a certain distribution of color straight When side's figure is interval, then the pixel quantity falling into this interval is added 1；Finally will fall into each again The pixel quantity of distribution of color histogram is respectively divided by image slices vegetarian refreshments sum, is returned One Color Distribution Features hist (x) changed, wherein x represents color histogram.

The Color Distribution Features value so come out is more accurate.

Such as Fig. 2, in another embodiment of the present invention, the acquisition of described grain distribution feature Method includes:

Step 301: convert the image into gray-scale map, obtains image L；RGB image is changed Having multiple method for gray-scale map, one of which is to utilize formula L=0.299*R+0.587*G+0.114*B changes, and wherein, R represents the redness of pixel and divides Amount, G represents the green component of pixel, and B represents the blue component of pixel.0.299、0.587、 0.114 is coefficient, and this coefficient is the most unique certainly, it is impossible to be interpreted as limitation of the present invention.

Step 302: with the template of a size of 3 pixel × 3 pixels, travels through described image L, To the LBP feature (i.e. textural characteristics) of each template, wherein obtain the side of template LBP feature Method includes:

The gray value of 9 pixels in note template is p_i(0≤i≤8), the wherein picture of template center Element gray value is designated as p₀；The gray value of other pixel in template is deducted p₀Obtain:

g_i=p_i-p₀,(1≤i≤8)；

To each calculated g_iCarry out binary conversion treatment: if g_i>=0 makes g_i=1, otherwise g_i=0；The gi value of the pixel that will be located in position i expands to 82 system numbers, obtains LBP (i) special Levy as, 1≤i≤8:

$\mathrm{LBP}\left(i\right)=\underset{q=0}{\overset{7}{\mathrm{\Σ}}}{g}_i\×2^q;$

The LBP feature of above-mentioned calculating can not tackle the requirement of invariable rotary, in order to obtain rotation not The LBP feature become, needs to perform further step 303: obtain the invariable rotary of each template LBP^riFeature；Wherein obtain the LBP of the invariable rotary of template^riThe method of feature includes:

To each LBP (i) of template by carrying out shifting function, 8 binary numbers can be obtained respectively According to, take a wherein minimum LBP as invariable rotary^ri(i) feature:

1≤i≤8 in formula, ROR represents shifting function, q Represent shift amount；

Step 304: add up the LBP of each invariable rotary in each template^riThe distribution feelings of (i) feature Condition: travel through the LBP of each invariable rotary of each template^riI () eigenvalue, statistics falls into each stricture of vagina The pixel quantity that reason distribution histogram is interval, then each grain distribution histogram will be fallen into Pixel quantity be respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature Hist (y), wherein y represents grain distribution histogram.

Here the histogrammic definition of grain distribution is similar with the definition of aforementioned distribution of color rectangular histogram.Stricture of vagina The textural characteristics of entire image is divided into some intervals by reason distribution histogram, then uses each pixel Situation in each interval distribution describes the ratio that different texture is shared in entire image.

When, after the Color Distribution Features obtaining image, calculating the Color Distribution Features phase of two width images Seemingly spend a detailed description of the invention of Sa, including:

Step 401: utilize formulaCalculate Color Distribution Features Similarity, wherein hist₁X () is the Color Distribution Features of piece image, hist₂X () is the second width The Color Distribution Features of image, wherein x represents color histogram.

When, after the grain distribution feature obtaining image, calculating the grain distribution feature phase of two width images Seemingly spend a detailed description of the invention of Sb, including:

Step 501: utilize formulaCalculate grain distribution feature phase Like degree, wherein hist₁Y () is the grain distribution feature of piece image, hist₂Y () is the second width figure The grain distribution feature of picture.Y represents grain distribution histogram.

Under the teaching of foregoing, those skilled in the art are readily apparent that and innovate based on the present invention A kind of similarity acquisition methods for grain distribution image search method of thought, including:

Step 1: extract the grain distribution feature of input picture；

Step 2: the grain distribution feature calculating described input picture respectively is each with data base The similarity of the grain distribution feature of width image, obtains input picture and each width figure in data base Grain distribution characteristic similarity Sb(i between Xiang), i takes 0,1,2 ... database images is total Number-1.

In like manner, special when the grain distribution obtained between input picture and the every piece image of data base After levying similarity, each similarity is ranked up, similarity the biggest explanation two width image more phase Seemingly, we can rule of thumb set a threshold value, will be greater than combining similarity more than this threshold value Image output in all data bases, as retrieval result.

The invention is not limited in aforesaid detailed description of the invention.The present invention expands to any at this In description disclose new feature or any new combination, and disclose arbitrary new method or The step of process or any new combination.

Claims (6)

1. for distribution of color and a similarity acquisition methods for grain distribution image retrieval, its It is characterised by, including:

Step 1: extract Color Distribution Features and the grain distribution feature of input picture；

Step 2: the Color Distribution Features calculating described input picture respectively is each with data base The similarity of the Color Distribution Features of width image, obtains input picture and each width figure in data base Color Distribution Features similarity Sa (i) between Xiang, i takes 0,1,2 ... database images is total Number-1；

Calculate respectively the grain distribution feature of described input picture with data base every piece image The similarity of grain distribution feature, obtains in input picture and data base between every piece image Grain distribution characteristic similarity Sb (i), i takes 0,1,2 ... database images sum-1；

Step 3: utilize formula S (i)=Wa × Sa (i)+Wb × Sb (i), i to take 0,1,2 ... Database images sum-1, Wa, Wb are weight coefficient and Wa+Wb=1, calculate input figure Combination similarity S (i) as piece image every with data base；

The method calculating Color Distribution Features similarity Sa in described step 2 includes:

Utilize formulaCalculate Color Distribution Features similarity, wherein hist₁X () is the Color Distribution Features of piece image, hist₂X () is that the color of the second width image is divided Cloth feature；

The method calculating grain distribution characteristic similarity Sb in described step 2 includes:

Utilize formulaCalculate grain distribution characteristic similarity, wherein hist₁Y () is the grain distribution feature of piece image, hist₂Y () is that the texture of the second width image divides Cloth feature.

The most according to claim 1 for distribution of color with the phase of grain distribution image retrieval Seemingly spend acquisition methods, it is characterised in that the acquisition methods of described Color Distribution Features includes:

Step 201: image is transformed into hsv color space, obtains image I；

Step 202: the H of each pixel of image, S, V component are mapped as color feature value G: G=Q_s*Q_v*H+Q_v*S+V；The span of three passages in hsv color space is carried out Interval division, is respectively divided into H_i,S_j,V_k, wherein 0≤i≤Q_h,0≤j≤Q_s,0≤k≤Q_v, Q_h,Q_s,Q_vRepresent the divided interval sum of three passages in hsv color space respectively；

Step 203: the color feature value distribution situation of each pixel in statistical picture: traversal The color feature value of each pixel, statistics falls into the pixel of each distribution of color histogram Point quantity, is respectively divided by image by the pixel quantity falling into each distribution of color histogram Pixel sum, obtains normalized Color Distribution Features hist (x), and wherein x represents distribution of color Histogram.

The most according to claim 2 for distribution of color with the phase of grain distribution image retrieval Seemingly spend acquisition methods, it is characterised in that the acquisition methods of described Color Distribution Features also includes:

Divide an image into N block；In described step 203: each pixel in statistical picture Eigenvalue distribution situation: traveling through the eigenvalue of each pixel, statistics falls into each color and divides The pixel quantity of cloth histogram, and will not be the pixel statistics two in image boundary block Secondary；The pixel quantity falling into each distribution of color histogram is respectively divided by image pixel Point sum, obtains normalized Color Distribution Features hist (x), and wherein x represents color histogram.

The most according to claim 1 and 2 for distribution of color with grain distribution image retrieval Similarity acquisition methods, it is characterised in that the acquisition methods of described grain distribution feature includes:

Step 301: convert the image into gray-scale map, obtains image L；

Step 302: with the template of a size of 3 pixel × 3 pixels, travels through described image L, To the LBP feature of each template, the method wherein obtaining template LBP feature includes:

The gray value of 9 pixels in note template is p_i(0≤i≤8), the wherein picture of template center Element gray value is designated as p₀；The gray value of other pixel in template is deducted p₀Obtain:

g_i=p_i-p₀,(1≤i≤8)；

To each calculated g_iCarry out binary conversion treatment: if g_i>=0 makes g_i=1, otherwise g_i=0；Will be located in the g of the pixel of position i_iValue expands to 82 system numbers, obtains LBP (i) special Levy as, 1≤i≤8:

$LBP\left(i\right)=\underset{q=0}{\overset{7}{\Σ}}{g}_i\×2^q;$

Step 303: obtain the LBP of the invariable rotary of each template^riFeature；Wherein obtain template The LBP of invariable rotary^riThe method of feature includes:

To each LBP (i) of template by carrying out shifting function, 8 binary systems can be obtained respectively Data, take a wherein minimum LBP as invariable rotary^ri(i) feature:

1≤i≤8 in formula, ROR represents shifting function, q Represent shift amount；

Step 304: add up the LBP of each invariable rotary in each template^riThe distribution feelings of (i) feature Condition: travel through the LBP of each invariable rotary of each template^riI () eigenvalue, statistics falls into each stricture of vagina The pixel quantity that reason distribution histogram is interval, then each grain distribution histogram will be fallen into Pixel quantity be respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature Hist (y), wherein y represents grain distribution histogram.

The most according to claim 1 for distribution of color with the phase of grain distribution image retrieval Seemingly spend acquisition methods, it is characterised in that described Wa > Wb.

6., for a similarity acquisition methods for grain distribution image search method, its feature exists In, including:

Step 1: extract the grain distribution feature of input picture；

Step 2: the grain distribution feature calculating described input picture respectively is each with data base The similarity of the grain distribution feature of width image, the some grain distribution characteristic similarity Sb arrived I (), i takes 0,1,2 ... database images sum-1；

The acquisition methods of described grain distribution feature includes:

Step 301: convert the image into gray-scale map, obtains image L；

Step 302: with the template of a size of 3 pixel × 3 pixels, travels through described image L, To the LBP feature of each template, the method wherein obtaining template LBP feature includes:

The gray value of 9 pixels in note template is p_i(0≤i≤8), the wherein picture of template center Element gray value is designated as p₀；The gray value of other pixel in template is deducted p₀Obtain:

g_i=p_i-p₀,(1≤i≤8)；

To each calculated g_iCarry out binary conversion treatment: if g_i>=0 makes g_i=1, otherwise g_i=0；Will be located in the g of the pixel of position i_iValue expands to 82 system numbers, obtains LBP (i) special Levy as, 1≤i≤8:

$LBP\left(i\right)=\underset{q=0}{\overset{7}{\Σ}}{g}_i\×2^q;$

Step 303: obtain the LBP of the invariable rotary of each template^riFeature；Wherein obtain template The LBP of invariable rotary^riThe method of feature includes:

To each LBP (i) of template by carrying out shifting function, 8 binary systems can be obtained respectively Data, take a wherein minimum LBP as invariable rotary^ri(i) feature:

1≤i≤8 in formula, ROR represents shifting function, q Represent shift amount；

Step 304: add up the LBP of each invariable rotary in each template^riThe distribution feelings of (i) feature Condition: travel through the LBP of each invariable rotary of each template^riI () eigenvalue, statistics falls into each stricture of vagina The pixel quantity that reason distribution histogram is interval, then each grain distribution histogram will be fallen into Pixel quantity be respectively divided by image slices vegetarian refreshments sum, obtain normalized grain distribution feature Hist (y), wherein y represents grain distribution histogram；

The method calculating grain distribution characteristic similarity Sb in described step 2 includes:

Utilize formulaCalculate grain distribution characteristic similarity, wherein hist₁Y () is the grain distribution feature of piece image, hist₂Y () is that the texture of the second width image divides Cloth feature.