DXMPsych2012Project - Revision history

imported>Psych2012: /* Conclusions */

2012-03-22T03:30:49Z

Conclusions

← Older revision		Revision as of 03:30, 22 March 2012
Line 95:		Line 95:
	= Conclusions =		= Conclusions =

	From the results of this project, we conclude that human skin classifiers in YCbCr and CIELAB color spaces perform better than skin classifier in HSV color space. However, the difference in performance between classifiers in different color spaces is not very ~~big~~. As a result, in a ~~application that used~~ skin-detection, performance different in different color spaces should not be the biggest factor in deciding which color space to use. The engineer who designs the application should consider to work in device output format to safe computation cost.		From the results of this project, we conclude that human skin classifiers in YCbCr and CIELAB color spaces perform better than skin classifier in HSV color space. However, the difference in performance between classifiers in different color spaces is not very significant. As a result, when designing a skin-detection software, performance different in different color spaces should not be the biggest factor in deciding which color space to use. The engineer who designs the application should consider to work in device output format to safe computation cost.
	We observed that ~~combine~~ classifiers have lower false dismissal rate than single-color-space classifiers. In addition, we confirmed the fact that false detection rate is higher for all classifiers, this error is a result of metamorphism. Because of high false detection rate, we have to do some post processing, such as applying morphological filters and connected component noise removal.		We observed that hybrid classifiers have lower false dismissal rate than single-color-space classifiers. In addition, we confirmed the fact that false detection rate is higher for all classifiers, this error is a result of metamorphism. Because of high false detection rate, we have to do some post processing, such as applying morphological filters and connected component noise removal.
	In this project, we implemented pixel based human skin classifiers. The results show that the performance of this kind of classifiers is far from ideal. The next step is to implement super-pixel based classifiers in order to improve the classification performance.		In this project, we implemented pixel based human skin classifiers. The results show that the performance of this kind of classifiers is far from ideal. The next step is to implement super-pixel based classifiers in order to improve the classification performance.

imported>Psych2012: /* Results with combined method */

2012-03-22T03:26:57Z

Results with combined method

← Older revision		Revision as of 03:26, 22 March 2012
Line 87:		Line 87:
	[[File:HybridClassifierErrorRate.jpg \|thumb\|upright=4\|center\| Figure 5: Error Plot of Hybrid Classifiers]]		[[File:HybridClassifierErrorRate.jpg \|thumb\|upright=4\|center\| Figure 5: Error Plot of Hybrid Classifiers]]

	Again, the false detection rate ~~are~~ high across the hybrid classifiers. The false detection rate, however, ~~are better~~ than single-color-space classifiers. "-HSV" hybrid classifier achieved a false dismissal rate of 0.09. Figure 6 shows an example of a test image classified by "-HSV" hybrid classifier. The classified image is filtered with morphological filters to remove some of the false positives.		Again, the false detection rate is high across the hybrid classifiers. The false detection rate, however, is lower than any of the single-color-space classifiers. "-HSV" hybrid classifier achieved a false dismissal rate of 0.09. Figure 6 shows an example of a test image classified by "-HSV" hybrid classifier. The classified image is filtered with morphological filters to remove some of the false positives.

imported>Psych2012: /* Results with combined method */

2012-03-22T03:25:17Z

Results with combined method

← Older revision		Revision as of 03:25, 22 March 2012
Line 83:		Line 83:

	== Results with combined method ==		== Results with combined method ==
	Figure 5 presents the error plot of hybrid classifiers. The first five hybrid classifiers combined results from 4 single-color-space classifiers. The name of the classifiers indicates which single-color-space classifier it did not use. For example, hybrid classifier "-RGB" combined results from <math>C_{CbCr}</math>, <math>C_{hs}</math>, <math>C_{xy}</math>, and <math>C_{ab}</math>, but it did '''not''' use the results from <math>C_{ab}</math>. The last hybrid classifier used results from all 5 single-color-space classifiers.		Figure 5 presents the error plot of hybrid classifiers. The first five hybrid classifiers combined results from 4 single-color-space classifiers. The name of the classifiers indicates which single-color-space classifier it did not use. For example, hybrid classifier "-RGB" combined results from <math>C_{CbCr}</math>, <math>C_{hs}</math>, <math>C_{xy}</math>, and <math>C_{ab}</math>, but it did '''not''' use the results from <math>C_{ab}</math>. The last hybrid classifier used results from all five single-color-space classifiers.
	<br>		<br>
	[[File:HybridClassifierErrorRate.jpg \|thumb\|upright=4\|center\| Figure 5: Error Plot of Hybrid Classifiers]]		[[File:HybridClassifierErrorRate.jpg \|thumb\|upright=4\|center\| Figure 5: Error Plot of Hybrid Classifiers]]

imported>Psych2012: /* Results of single-color-space classifiers */

2012-03-21T23:53:25Z

Results of single-color-space classifiers

← Older revision		Revision as of 23:53, 21 March 2012
Line 80:		Line 80:
	[[File:SingleColorSpaceErrorRate.jpg \|thumb\|700px\|center\| Figure 4: Error Plot of Single-Color-Space Classifiers]]		[[File:SingleColorSpaceErrorRate.jpg \|thumb\|700px\|center\| Figure 4: Error Plot of Single-Color-Space Classifiers]]

	First, we observed that false detection rate ~~are~~ higher than false dismissal rate for all classifiers. This is expected because of metamerism. We also noticed that the false dismissal rate and false detection rate of <math>C_{CbCr}</Math> and <math>C_{ab}</Math> are lower than the error rates of <math>C_{sv}</math>. This result indicates that skin detection performed in YCbCr and CIELAB color spaces is more accurate than skin detection in HSV color space.		First, we observed that false detection rate is higher than false dismissal rate for all classifiers. This is expected because of metamerism. We also noticed that the false dismissal rate and false detection rate of <math>C_{CbCr}</Math> and <math>C_{ab}</Math> are lower than the error rates of <math>C_{sv}</math>. This result indicates that skin detection performed in YCbCr and CIELAB color spaces is more accurate than skin detection in HSV color space.

	== Results with combined method ==		== Results with combined method ==

imported>Psych2012: /* Hybrid Method */

2012-03-21T23:49:55Z

Hybrid Method

← Older revision		Revision as of 23:49, 21 March 2012
Line 71:		Line 71:

	===Hybrid Method===		===Hybrid Method===
	We also constructed a hybrid ~~classifier~~ that combines the results from multiple single-color-space classifiers. The hybrid ~~classifier~~ used a majority vote approach. For example, if we use <math>C_{rg}</math>, <math>C_{CbCr}</math>, <math>C_{xy}</math>, and <math>C_{ab}</math> in our hybrid classifier, then a pixel is labeled 1 if 2 or more single-color-space classifiers labeled the pixel to be 1. We evaluated the performance of hybrid ~~classifier~~ with different combination of single-color-space classifiers.		We also constructed hybrid classifiers that combines the results from multiple single-color-space classifiers. The hybrid classifiers used a majority vote approach. For example, if we use <math>C_{rg}</math>, <math>C_{CbCr}</math>, <math>C_{xy}</math>, and <math>C_{ab}</math> in our hybrid classifier, then a pixel is labeled 1 if two or more single-color-space classifiers labeled the pixel to be 1. We evaluated the performance of hybrid classifiers with different combination of single-color-space classifiers.

	= Results =		= Results =

imported>Psych2012: /* Classifier */

2012-03-21T23:43:05Z

Classifier

← Older revision		Revision as of 23:43, 21 March 2012
Line 52:		Line 52:

	===Classifier===		===Classifier===
	The classifier for a color space is a binary matrix that has the same size as the corresponding histogram. Let <math> C_{rg} </math> be the classifier of RGB color space, and let <math> T_{rg} </math> be the threshold for <math>H_{rg}</math>. Then		The skin classifier for a color space is a binary matrix that has the same size as the corresponding histogram. Let <math> C_{rg} </math> be the classifier of RGB color space, and let <math> T_{rg} </math> be the threshold for <math>H_{rg}</math>. Then
	:<math> C_{rg}[i,j] = \mathbf{1}\{H_{rg}[i,j] > T_{rg}\} </math>		:<math> C_{rg}[i,j] = \mathbf{1}\{H_{rg}[i,j] > T_{rg}\} </math>
	~~classifiers~~ for other color spaces <math>C_{CbCr}</math>, <math>C_{HS}</math>, <math>C_{xy}</math> and <math>C_{ab}</math> are similarly constructed. Figure 3 shows the trained classifiers of the 5 color spaces.		Classifiers for other color spaces <math>C_{CbCr}</math>, <math>C_{HS}</math>, <math>C_{xy}</math> and <math>C_{ab}</math> are similarly constructed. Figure 3 shows the trained classifiers of the 5 color spaces.

	We define 1 to indicate skin, and 0 to indicate non-skin. A pixel <math>I</math> with chromatic features <math>[r \; g \; Cb \; Cr \; H \; S \; x \; y \; a \; b]</math> is labeled either 1 or 0 with following equations. We only present classification using RGB classifier <math>C_{rg}</math>, since classification using classifiers in other color spaces are applied in a similar way.		We define 1 to indicate skin, and 0 to indicate non-skin. A pixel <math>I</math> with chromatic features <math>[r \; g \; Cb \; Cr \; H \; S \; x \; y \; a \; b]</math> is labeled as either 1 or 0 with following equations. We only present classification using RGB classifier <math>C_{rg}</math>, since classification using classifiers in other color spaces are applied in a similar way.
	:<math> \quad I_{label} = C_{rg}[i,j] </math>		:<math> \quad I_{label} = C_{rg}[i,j] </math>
	:<math> i = floor(\frac{r}{\tau}) \quad j = ceiling(\frac{g}{\tau})</math>		:<math> i = floor(\frac{r}{\tau}) \quad j = ceiling(\frac{g}{\tau})</math>

imported>Psych2012: /* Histogram */

2012-03-21T23:36:02Z

Histogram

← Older revision		Revision as of 23:36, 21 March 2012
Line 42:		Line 42:
	:<math> H_{rg}[i,j] = \frac{max(H_{rg}(i,j), 0)}{max(H_{rg})} </math>		:<math> H_{rg}[i,j] = \frac{max(H_{rg}(i,j), 0)}{max(H_{rg})} </math>
	The histogram is further smoother by a 2D Gaussian filter <math>G_{\sigma}</math> whose variance is <math>\sigma ^2</math>.		The histogram is further smoother by a 2D Gaussian filter <math>G_{\sigma}</math> whose variance is <math>\sigma ^2</math>.
	:<math> H_{rg}[i,j] = (H_{rg} * G_\sigma)[i,j] </math>		:<math> H_{rg}[i,j] = (H_{rg} \star G_\sigma)[i,j] </math>
	Histograms for other color spaces <math>H_{CbCr}</math>, <math>H_{HS}</math>, <math>H_{xy}</math> and <math>H_{ab}</math> are similarly constructed.		Histograms for other color spaces <math>H_{CbCr}</math>, <math>H_{HS}</math>, <math>H_{xy}</math> and <math>H_{ab}</math> are similarly constructed.

imported>Psych2012: /* Histogram */

2012-03-21T23:34:11Z

Histogram

← Older revision		Revision as of 23:34, 21 March 2012
Line 42:		Line 42:
	:<math> H_{rg}[i,j] = \frac{max(H_{rg}(i,j), 0)}{max(H_{rg})} </math>		:<math> H_{rg}[i,j] = \frac{max(H_{rg}(i,j), 0)}{max(H_{rg})} </math>
	The histogram is further smoother by a 2D Gaussian filter <math>G_{\sigma}</math> whose variance is <math>\sigma ^2</math>.		The histogram is further smoother by a 2D Gaussian filter <math>G_{\sigma}</math> whose variance is <math>\sigma ^2</math>.
	:<math> H_{rg}[i,j] = (H_{rg} ~~\ast~~ G_\sigma)[i,j] </math>		:<math> H_{rg}[i,j] = (H_{rg} * G_\sigma)[i,j] </math>
	Histograms for other color spaces <math>H_{CbCr}</math>, <math>H_{HS}</math>, <math>H_{xy}</math> and <math>H_{ab}</math> are similarly constructed.		Histograms for other color spaces <math>H_{CbCr}</math>, <math>H_{HS}</math>, <math>H_{xy}</math> and <math>H_{ab}</math> are similarly constructed.

imported>Psych2012: /* Histogram */

2012-03-21T23:33:25Z

Histogram

← Older revision		Revision as of 23:33, 21 March 2012
Line 42:		Line 42:
	:<math> H_{rg}[i,j] = \frac{max(H_{rg}(i,j), 0)}{max(H_{rg})} </math>		:<math> H_{rg}[i,j] = \frac{max(H_{rg}(i,j), 0)}{max(H_{rg})} </math>
	The histogram is further smoother by a 2D Gaussian filter <math>G_{\sigma}</math> whose variance is <math>\sigma ^2</math>.		The histogram is further smoother by a 2D Gaussian filter <math>G_{\sigma}</math> whose variance is <math>\sigma ^2</math>.
	:<math> H_{rg}[i,j] = (H_{rg} \~~star~~ G_\sigma)[i,j] </math>		:<math> H_{rg}[i,j] = (H_{rg} \ast G_\sigma)[i,j] </math>
	Histograms for other color spaces <math>H_{CbCr}</math>, <math>H_{HS}</math>, <math>H_{xy}</math> and <math>H_{ab}</math> are similarly constructed.		Histograms for other color spaces <math>H_{CbCr}</math>, <math>H_{HS}</math>, <math>H_{xy}</math> and <math>H_{ab}</math> are similarly constructed.

imported>Psych2012: /* Histogram */

2012-03-21T23:31:31Z

Histogram

← Older revision		Revision as of 23:31, 21 March 2012
Line 37:		Line 37:
	Let <math>H_{rg}</math> be the histogram of RGB color space, and <math>H_{rg}(i,j)</math> is the value of bin <math>[i,j]</math> of the histogram. Let <math>N(k)</math> be the number of pixels in training image <math> k </math>, and <math>I_{(k, l)}</math> be the <math>l^{th}</math> pixel of <math>k^{th}</math> image. Define <math>r_{(k, l)}</math> and <math>g_{(k, l)}</math> to be the <math>r</math> and <math>g</math> features of <math>I_{(k, l)}</math>. In addition, we denote the bin size of the histogram as <math>\tau</math>. The histogram can be expressed as		Let <math>H_{rg}</math> be the histogram of RGB color space, and <math>H_{rg}(i,j)</math> is the value of bin <math>[i,j]</math> of the histogram. Let <math>N(k)</math> be the number of pixels in training image <math> k </math>, and <math>I_{(k, l)}</math> be the <math>l^{th}</math> pixel of <math>k^{th}</math> image. Define <math>r_{(k, l)}</math> and <math>g_{(k, l)}</math> to be the <math>r</math> and <math>g</math> features of <math>I_{(k, l)}</math>. In addition, we denote the bin size of the histogram as <math>\tau</math>. The histogram can be expressed as
	:<math>H_{rg}[i,j] = \sum\limits_{k=1}^{100}\sum\limits_{l=1}^{N(k)}\mathbf{1}\{i\tau \leq r_{k,l} < (i+1)\tau\} \mathbf{1}\{ j\tau \leq g_{k,l} < (j+1)\tau\} (\mathbf{1}\{isskin(I_{k,l})\} - \mathbf{1}\{\lnot isskin(I_{k,l})\}) </math>		:<math>H_{rg}[i,j] = \sum\limits_{k=1}^{100}\sum\limits_{l=1}^{N(k)}\mathbf{1}\{i\tau \leq r_{k,l} < (i+1)\tau\} \mathbf{1}\{ j\tau \leq g_{k,l} < (j+1)\tau\} (\mathbf{1}\{isskin(I_{k,l})\} - \mathbf{1}\{\lnot isskin(I_{k,l})\}) </math>
	In other word, <math>H_{rg}(i,j)</math> is the number of skin pixels minus the number of non-skin pixels in the range ~~that is specified by~~ bin <math>)(i,j)</math>.		In other word, <math>H_{rg}(i,j)</math> is the number of skin pixels minus the number of non-skin pixels in the range of bin <math>(i,j)</math>.

	We then normalize the histogram to <math>[0, 1]</math> by		We then normalize the histogram to <math>[0, 1]</math> by