About Perception and Hue Histograms in HSV Space

Federal University of Rio de Janeiro (UFRJ) :: Campus Cidade Universitária :: Rio de Janeiro :: Fundão Island :: COPPE/PESC/Laboratory of Computer Graphics :: March/2015

Michel Alves -- Five Minute Speech (FMS) :: Overview of Activities Developed in Disciplines and Guided Studies :: Graduate Program in Systems Engineering and Computing

About Perception and HueAbout Perception and HueHistograms in HSV SpaceHistograms in HSV Space

Michel Alves dos SantosMay, 2015

*original

*hue-saturation projection

*saturation

*brightness*hue histogram



IntroductionIntroduction

red channel green channel blue channel

What's your perception aboutthe colors of this image?

*Barns Grand Teton: Grand Teton National Park

Note the marked areas!

And the distribution by channel?How each color component behaves in its proper channel?



IntroductionIntroduction*Barns Grand Teton: Grand Teton National Park

Observe the distribution ofenergy for each channel:

But, about the hues? What the RGB model can tell us? This kind of graph shows us how the energy of each channel is distributed!

red channel green channel blue channel



IntroductionIntroduction*Barns Grand Teton: Grand Teton National Park

Firstly, we will use the HSV Color Space ...

How to build a hue histogram?

Now, note the diagram below:

Can you identify hues distinctly?

Probably not! Because of clustersand the intersection of colors!

A histogram of hues can help usimprove our perception!



RGB Vs HSV RGB Vs HSV (Color Spaces)(Color Spaces)

RGB Color Space HSV Color Space*Solids produced using POVRay [http://lib.povray.org/]

Simplicity in representation and storage Close to human perception

Some point operations are faster... Easy interpolation between colors

Invariant to illumination transformsHave lighting and color properties togetherHave lighting and color properties together



RGB Vs HSV RGB Vs HSV (Color Spaces – a common mistake...)(Color Spaces – a common mistake...)


A cone is commonly adopted but cylindricalrepresentation is more mathematically accurate.



RGB Vs HSV RGB Vs HSV (Color Spaces – Solids Transition)(Color Spaces – Solids Transition)




RGB Vs HSV RGB Vs HSV (Transformation)(Transformation)

RGB HSV Transformationisn't linear!

Some care should be takenwith monochromatic

pixels!*[avoid divisions by zero]



RGB Vs HSV RGB Vs HSV (Transformation)(Transformation)

RGB HSV Transformationisn't linear!

Alternative

Some care should be takenwith monochromatic

pixels!*[avoid divisions by zero]



RGB Vs HSV RGB Vs HSV (Color Image Thresholding)(Color Image Thresholding)

Find the yellow M&M's!

A simple task for a humanobserver, but not so easy

for machines:

Complexity depends of the space!

Using RGB:R:

G:

B:

200 255

200 255

0 50

Using HSV:(3 intervals) (1 interval)




Find the yellow M&M's!

A simple task for a humanobserver, but not so easy

for machines:

Complexity depends of the space!

Using RGB:R:

G:

B:

200 255

200 255

0 50

Using HSV:(3 intervals) (1 interval)

[This approach is independent of saturation or brightness]





*hues of sky *hues of barn *hues of grass

Using this feature some taskscan be easily performed!

As detection of regions or objectsby its characteristic colors!It's one of the bases of color perception!



RGB Vs HSV RGB Vs HSV (Color Perception)(Color Perception)

And now, what's your perceptionabout the colors of this image?


Where we have more blue hues?

Where we have more orange hues?

Where we have more green hues?

Here we have a blend of yellow and green hues!

Do not fool yourself!

(Grass???)

(Sky)

(Barn)

*hue histogram

*hue projection



RGB Vs HSV RGB Vs HSV (Color Perception)(Color Perception)

And now, what's your perceptionabout the colors of this image?


Where we have more blue hues?

Where we have more orange hues?

Where we have more green hues?

Here we have a blend of yellow and green hues!

Do not fool yourself!

(Grass???)

(Sky)

(Barn)

*hue histogram

*hue projection



RGB Vs HSV RGB Vs HSV (Meanings of the Components)(Meanings of the Components)

Energy of red channel Energy of green channel Energy of blue channel

What is the basic color? How vibrant is the color? How brightness is the color?



Hue HistogramsHue Histograms

*hue projection *saturation *value

We will build our histograms takingaccount the number of positions

of the hue circle!Number of bins (discrete positions): 0..360 or 0 to 2π

Note that the hue channel is a ring structure!



Hue Histograms - StepsHue Histograms - Steps

Our input images are in RGB Color Space, the output is a transformed hue histogram...

We are applying a square root transformation to obtain the final hue histogram...

Observe the followdiagram!

The metric of contribution is a flexible part of the process and depends of application...



Hue Histograms - StepsHue Histograms - Steps

Our input images are in RGB Color Space, the output is a transformed hue histogram...

We are applying a square root transformation to obtain the final hue histogram...

Observe the followdiagram!

The metric of contribution is a flexible part of the process and depends of application...

Get to work!

Now, I will showsome metrics!



Hue Histograms :: Simple FrequencyHue Histograms :: Simple Frequency*Barns Grand Teton: Grand Teton National Park

Kronecker Delta

Hue

Picture or Image

*hue histogram

This is a simple counting!

It doesn't consider otherHSV channels.

Bin Position

PixelHistogram

Notation



Hue Histograms :: by SaturationHue Histograms :: by Saturation*Barns Grand Teton: Grand Teton National Park

*hue histogram

Kronecker Delta

Hue

Picture

Only considers the saturation channel...

Saturation

Ref: [LP'15]

Bin Position

PixelHistogram



Hue Histograms :: by ContributionHue Histograms :: by Contribution*Barns Grand Teton: Grand Teton National Park

*hue histogram

Kronecker DeltaSaturation ValuePicture

HueHistogram Pixel

Bin Position



Hue Histograms :: Hue Histograms :: Weighted by ContributionWeighted by Contribution*Barns Grand Teton: Grand Teton National Park

*hue histogram

Kronecker DeltaSaturation ValuePicture

Hue

*similar to the previous!

Normalization FactorBin Position

Pixel



Hue Histograms :: Hasan WeightedHue Histograms :: Hasan Weighted Ref: [MAH'12]

*hue histogram


Picture Previous Histogram

*exponential weighted histogram

Mean of Hues

*similar to the previous???

Mean

Variance

Variance of Hues

Notation

Weighted Histogram

Bin Position



Hue Histograms :: Hasan WeightedHue Histograms :: Hasan Weighted Ref: [MAH'12]

*hue histogram



*exponential weighted histogram

Mean of Hues

*similar to the previous???

Variance of Hues *before weighting *after weighting

frequency frequencyhue hue

Weighted Histogram

Bin Position



Hue Histograms :: Gaussian WeightedHue Histograms :: Gaussian Weighted

Gaussian Function


*hue histogram


Weighted Histogram

Bin Position

Mean of Hues

Variance of Hues



Hue Histograms :: Hue Histograms :: Smoothed by BoxSmoothed by Box*hue histogram, window size = 1x9


Box function

Convolution

Gaussian Function

Previous HistogramWeighted Histogram



Hue Histograms :: Hue Histograms :: Smoothed by BoxSmoothed by Box*hue histogram, window size = 1x9


Box function

Convolution

Gaussian Function


*smoothing employing a boxcar function



Hue Histograms :: Hue Histograms :: Smoothed by GaussianSmoothed by Gaussian*hue histogram, window size = 1x9


Gaussian function

Convolution

Gaussian Function




Hue Histograms :: Hue Histograms :: Smoothed by a KernelSmoothed by a Kernel*hue histogram, window size = 1x11


kernel

Kernel function

Convolution

Gaussian Function




Hue Histograms :: Hue Histograms :: Smoothed by a KernelSmoothed by a Kernel*hue histogram, window size = 1x11


kernel function employed: Epanechnikov

kernel

Kernel function

Convolution

Gaussian Function




Hue Histograms ::Hue Histograms :: Spatial Color Coherence Spatial Color Coherence*hue histogram, window size = 1x11


kernelSpatial Color Coherence Neighborhood Difference

Kernel functionConvolution

*8-connected neighborhood

Gaussian FunctionWeighted Histogram Saturation Value Kronecker Delta



Hue Histograms ::Hue Histograms :: Spatial Color Coherence Spatial Color Coherence*hue histogram, window size = 1x11


kernelSpatial Color Coherence Neighborhood Difference

Kernel functionConvolution

*8-connected neighborhood

Gaussian FunctionWeighted Histogram Saturation Value Kronecker Delta

*Spatial Color Coherence Map



Hue Histograms :: The end of our jorney...Hue Histograms :: The end of our jorney...Simple Frequency

By Contribution

Weighted Contribution

Hasan Weighted

By Saturation

Gaussian WeightedSmoothed

A brief visualA brief visualcomparison!comparison!

This is the end!



ConclusionConclusion

Use the appropriate space for your problem! Always!

Hue based histograms can be employed in a wide range of applications!

Color spaces that have hue separation approximate human perception!

Some considerations should be taken into account in the construction of a huehistogram as brightness and perceptive contribution...



Thanks for your attention!Thanks for your attention!Michel Alves dos SantosMichel Alves dos SantosReferences:References:

Digital Image Processing (3rd Ed), Gonzalez & Woods, Pages 407-437. [GW'08]

Blind Stain Decomposition for Histo-Pathology Images Using Circular Natureof Chroma Components, Li & Plataniotis, 2015, Page 2. [LP'15]

Color Image Segmentation Using Acceptable Histogram Segmentation, Delonet al., 2005, LNCS.

Shot Classification Using Domain Specific Features for Movie Management, Hasan et al., 2012, LNCS, Springer. [MAH'12]

Contrast Enhancement of Color Images Based on Wavelet Transform and Human Visual System, 2007, Xiao & Ohya, IASTED.

Progressive Histogram Reshaping for Creative Color Transfer and ToneReproduction, Pouli & Reinhard, 2010, NPAR2010.

About Perception and Hue Histograms in HSV Space

Education

janeiro ufrj

federal university of

march2015 michel alves

proper channel

campus cidade universitria

barns grand teton

systems engineering

graduate program