Detecting skin in colour images, Computer Graphics

In this lab you will learn how to use chrominance1 to segment coloured images. Here you be detecting skin, however, you could use this method to detect other coloured regions in images.

Firstly you will build a model describing the likelihood that any pair of chrominance values belongs to a piece of skin. Secondly you will test images and identify likely skin regions.

Getting Started

1. Download the following files from the course web site

  • the image 'face.jpg'.
  • the incomplete file 'Lab2.m', and
  • the file 'RGB2Lab.m'

Save these in the directory you are using for this lab.

Deliverable:

To gain the marks for this lab you will need to show me your completed Lab2 function running during the lab. Complete the function Lab2 by writing the two sub-functions

make_chroma_model, and

find_chroma

The specifications of these functions can be found in the downloadable file Lab2.m

What to do

Firstly have a look at the Lab3 file. Run it and see what it does. First it defines some constants num_bins and gsize. Then it loads an image and converts it to double in the range [0,1], but leaves it in RGB format. It then crops a region defined by the user.

While you write and debugging this program it is convenient not to have to select a region each time you run the program. Comment out the lines that display the prompt to the user and interactively crop the image, we'll enable them again later once the code's working. For the meantime crop the image automatically like this:

sample_colour = im_RGB(130:150,90:138,:);

This takes rows 130 to 150, columns 90 to 138, and all 3 colour channels. Notice that the

':' is used to mean 'everything inbetween' when it's between two indexes, or 'everything' when it's on it's own.

1 Colour has 3 channels, eg RGB, HSV. Chrominance is a two channel derivative of a colour that is independent of intensity. See lecture on Colour theory.

2 Now let's start on the function make_chroma_model. Firstly we need to convert from RGB colour space to CIE Lab colour space using RGB2Lab.m,

[L,a_chroma,b_chroma] = RGB2Lab(sample_colour);

RGB2Lab returns real a,b chrominance values in the range [-120,120] . We want to use these as indexes to an accumulator array, so we need to convert them to integers in the range [0, num_bins], where num_bins is a constant defining the size of each dimension of our accumulator array. To do this we will write a very short sub-function.

Write a sub-function ab2ind that is passed 2 parameters ab_chroma and num_bins and returns an index ind. This will only take two lines. Firstly convert ab_chroma from the range [-120,120] to [0,1].

ab_01 = ((ab_chroma)+120)/240;

Secondly discretise this into an integer index ind in the range [1,num_bins].

ind = round(ab_01*(num_bins-1)) + 1;

round rounds to the nearest integer. Note the minus 1 and plus 1 that are necessary to firstly scale to the range [0, num_bins-1] then increase this to [1, num_bins].

Getting back to make_chroma_model, we can now call our new sub-function to convert our a,b chrominance values to indexes for the accumulator array, e.g. for

a_chroma

a_ind = ab2ind(a_chroma, num_bins);

Now we are ready to start building the skin chrominance model. Create an accumulator array, this will be a num_bins × num_bins matrix of zeroes, use the zeros function to do this, call this matrix accum_array. We will use this for counting the occurrences of different chroma pairs and it will form the basis of our skin chrominance model.

Now we need to count the number of occurrences of each chroma pair. To do this you will need to use for loops to consider every pixel in the image. Look at the a find and b find values of each pixel and increment the appropriate cell in the accumulator array.

Done that? Have a look at your result using imagesc as follows, figure;

imagesc(accum_array); axis image;

xlabel('a'); ylabel('b'); title('Points in ab space');

 

Posted Date: 2/25/2013 7:33:38 AM | Location : United States







Related Discussions:- Detecting skin in colour images, Assignment Help, Ask Question on Detecting skin in colour images, Get Answer, Expert's Help, Detecting skin in colour images Discussions

Write discussion on Detecting skin in colour images
Your posts are moderated
Related Questions
Question 1: a. List respectively ten main technical and human components necessary to ensure a studio-based one-plus-one interview? b. How does each component contribute in

Variation of Intensity - Modeling and Rendering According to the phong model the variation of Intensity (I) along with α (since I α cos n α) is: i) for shiny surface (

Buffer Areas Required For Z-Buffer Algorithm For applying z-buffer algorithm, we need two buffer areas or two 2-Dimentional arrays: 1) Depth-buffer [i,j], to sa

List out the merits and demerits of Penetration methods?  The merits and demerits of the Penetration methods are as follows  It is a less expensive technique It has

For this assignment, you will add to the drawing program new features that are similar to features that you already have. Apply object-oriented design concepts such as inheritance

What is run length encoding?   Run length encoding is a compression method used to store the intensity values in the frame buffer, which keeps each scan line as a set of intege

Categorization of Light resources - Point source This is the easiest model for a light emitter. Currently rays from source obey radically diverging ways from the source positi

2-DIMENSIONAL TRANSFORMATION Modeling transformation is the mechanism to compose an image from modeling  primitives. The modeling primitives are defined in their own coordinate

Question : (a) List and describe important aspects to take into consideration when looking at paper for print. (b) Describe clearly the meaning of the following words: I.

Vertical retrace - Display Devices In a refresh CRT monitor, the time it takes for an electron beam to return to the top, left most point on the monitor after refreshing all ho