image processing and analysis i materials extracted from gonzalez & wood and castleman

Image Processing and Analysis I

Materials extracted from Gonzalez & Woodand Castleman

Light source

Detector

Intermediate Optics

Specimen

A typical biomedical optics experiment

Image Processing

Image Data BaseKnowledge Base

Digital Image ProcessingA process to extract information from image data

Image Acquisition

Image Annotation,

Compression,Storage/Retrieval

Preprocessing Segmentation Representation Recognition

Interpretation,Modeling

ProblemDefinition Results

Display,Animation

Materials extracted fromReferences texts:

Gonzalez & Woods, and

Castleman

Image Processing Example 1 – Pap Smear

Benign Squamous Cells Squamous Cell Carcinoma

One of the few histopathological tasks where image recognition system is becoming commercial

Image Processing Example 2 – Breast X-Ray

The distinction between benign and malignant can be difficult for breast x-rayRadiologist are highly trained in image recognition

Most biomedical imaging today does not address underlying molecular and cellular based mechanisms

Image Data Base – Format and Data Base

Header

Hyper-Data

Data

Monochrome Color

1 bit – binary

8 bit

16 bit

32 bit

32 bits (float)

RGBR (8 bit)G (8 bit)B (8 bit)

Gray

Scale

Image Preprocessing – histogram and contrast

Image Preprocessing – histogram equalization

Let r be the gray level value of a pixel in the image.

]1,0[r ; Map each gray level value r to a new value s: )(rTs The histogram distribution of the original image is: )(rPr

The histogram distribution of the new image is: )(sPs

In general: )(1])([)(sTrrs ds

drrpsP

Histogram equalization is defined as the transform: r

r dwwprTs0

)()(

Since )(rPdr

dsr , 1)( sPs for histogram equalization

Image Preprocessing – histogram and contrast

Convolution and Image Processing

Recall the definition of convolution:

dthgthtg )()()()(

Graphical explanation of convolution:

Convolution Theorem

)(~

)(~))(( fhfgfhg

dtedthgdtethg ftifti 22 )()()(

)(22 )()( tfifi etdthegd

)'(22 )'(')( tfifi ethdtegd

)(~

)(~ fhfg

where dtdttt ''

Image Preprocessing – Noise Reduction, averaging, low pass filter, median filter

1

1 M

M

Averaging M images reducenoise by sqrt(M)

Low Pass Filter

3x3 kernel 5x5 kernel

111

111

111

9

1

Median Filter

Replace center pixel value by the median value

from a nxn pixel neighorhood

11111

11111

11111

11111

11111

25

1

Avg = 30; Median =10

200108

1296

10127

Avg by 2,8,16,32,128 Kernel 3,5,7,15,25 5x5 low pass vs median