Image Processing

What are various image processing techniques?
1. Image Enhancement : Improves quality of image for human perception by removing noise blurring from image or video database.

2. Histogram based processing technique. :
Increasing Contrast using histogram and intensity techniques
3. Intensity Transformation Method. :
Spatial Filtering Transformation. Low pass and 2nd order filtering technique

What is segmentation method?
Segmentation method is used to partition the images into segment to find the region of interest for analysis of objects and others.

What is Feature Extraction?
Feature extraction- dimensionality reduction, represents interesting parts of an image as compact feature vector.

Feature Representation for --------.
(Compact Rep. of image data)

Where is feature extraction used?
When we have a large image.

What are various types of Feature extraction methods?
Slide 6.

What are two types of Face Recognition approaches?
1. Linear Subspace Method
2. Non-Linear Subspace Method

What all comes under Linear subspace method?
PCA, LDA, DCV

What all comes under non-linear subspace method?
KPCA, KLDA

What are feature invariant method?
structural features of object remain invariant when pose, view point or lighting condition changed.

What are the advantages of local features?
Locality, Quantity, Distinctiveness, Efficiency

What do you mean by locality?
Robust to occlusion and clutter.

Quantity means ------- in a single image.
(100s or 1000s)

Efficiency means ------- achievable.
(Real Time perf.)

What are appearance based method or holistic method?
Models/templates leaarned from a set of training images capture the compact representative variablity of facial appearances.

slide 17

What is PCA?
What is positive covariance and negative covariance?

What are Eigen vectors and eigen values?
Ax = lambdax
lambda = eigen value
Det(A-lambdaI) =0

Eigen Face Based Method?
1. Create matrix (A) fortraining image dataset.
2. Compute covariance matrix C from A
3. Compute eigen vector.
4. Select few most significant eigen vector forface recognition.
5.Compute coefficeint vector.
6. Coefficient value are used for cluster dataset and compute mean of cluster.

Advantage of eigen face method for face recognition.?
1. Raw intensity data used directly, no low/mid level proc.
2. No knowledge of geo./relectance required.
3. Data compression - lowDsubspace rep.
4. Recog. is simple and efficient

What are disadvantages of eigen face method?
1. Learning - Time consuming, difficult to update face data base
2. Appearance Based nature
3. No of face classes > D of the face space(only efficient)
4.It is very sensitive to scale
5. Problem when there is change in pose expression and disguise(used in frontal lobe)
6. All face image tested are taken against uniform background.

What does Eigen Faces method do?
Eigen Faces method finds linear comb. of features maximizes total variance in data.
In this discriminative info may be lost 'cuz it does not considers classes.

Eigenfaces are
the eigenvectors of the covariance matrix of
the probability distribution of
the vector space of
human faces

How to genrate set of Eigen Faces?
1. Large set of digitized images of human faces
2. Image normalized to line up eyes and mouth.
3. Eigen vectors of covarince matix extracted.
4. These eigen vectors are eigen faces.

Eigen Face summed together?
Eigen Face summed together -> Grayscale rendering of human face.

What type of method is LDA?
LDA is supervised learning method for classification and identification of face image.

What is the working principle of LDA?
Find feature vector in underlying space that best discriminates among st classes
For sample of all classes between class and within class scatter matrix is defined.

Goal of LDA?
Minimize within class scatter matrix
Maximize between class scatter matrix
Maximize det[SB]/det[SW]