Spatial Image Enhancement Flashcards
Image enhancement
- Transformations that highlight image features
- mathematical manipulations of data
- new data created from raw data
Fourier methods
- Performed in the frequency
9-point smoothing filter
- applied to all central pixels, not pixels around edges
- convolution kernal (mask) makes pixels more alike
Weighted smoothing filter
- convolution kernel is weighted
- pixel weight determined by proximity to central pixel
- less blurring and better spatial resolution compared to unweighted smoothing
- AKA replaces the averaging
- new pixel value= Σ(wt. x pixel value)/16(total wt. of kernal)
Edge enhancement filter
- opposite to image smoothing
- uses the same pixel replace averaging technique
- negative coefficients used in weighting
- kernel enhances edges, increasing contrast.
Mask Size
- The mask or kernel always has an odd number of rows and columns of pixels (edge enhancement and smoothing)
- Dimensions vary according to the equation (2n+1) x (2n+1)
- 3x3, 5x5, 7x7
Point processing operations
- Manipulations of the number of cts/pixel
- completely independant of neighbouring pixels
- background subtraction, gray scales, and color tables
Background subtraction
- Neighbourhood independant
- subtracting pre a determined, constant number of counts from each pixel in an image
- Improves contrast, but comes at the cost of increased noise
Interpolated background subtraction
- neighbourhood dependant
- does not assume that background is uniform, but varies with position in the image
- creates an ROI, all pixels outside ROI are considered background
- background pixels are weighed according to their distance from ROI
Gray scale
- neighbourhood independant, referred to also as the dynamic range
- number of shades between complete black and complete white
- translating the number of counts in a pixel to an integer, which in turn defines the pixels color
Linear Gray Scale
- Gray scale assigned is proportional to the number of counts in the pixel
- The pixel with the highest number of counts is assigned white and the pixel with no counts is assigned black
- all other pixels are given a gray scale proportional to the ratio of counts in each pixel to the maximum counts per pixel
- Normalizes pixel counts
Non-normalized gray scale
- Better suited for low count images
- causes loss of image contrast in high count images
- Limits dynamic range
- If hottest pixel is 200 counts, the gray level is 200
- all pixels are given the shade that corresponds to the absolute count/pixel value
Thresholding gray scale
- a variation of linear gray scale
- eliminates background
- cuts off lower percentage of maximum counts
- all pixels less than a set percentage of maximum counts will appear black and the gray scale will increase from there
Logarithmic and exponential gray scale
- grey scales can be logarithmic or exponential in design
- logarithmic enhance image contrast in low count areas and decreases it in high count areas
- exponential is opposide to logarithmic
- both should be used cautiously as they can result in false interpretation of RP distribution
color translation tables
- lookup tables used to assign colors or different intensities of the same color to each pixel as a function of pixel counts
- thre intensity scales are used: red, green, blue
- user preference
Quantitative analysis
- ROIs
- curve generation and analysis
Types of ROI
- manual
- semi-automated
- automated
manual ROIs
visual determination with manual use of mouse to draw ROI
Semi- automated ROIs
Tech has some input to alter the ROI
Automated ROIs
- Tech shows computer general area
- computer uses edge detection to determine ROI
Time activity curves
Determine flow characterisitices of an RP in the body
Issues with TACs
- Noise interference
- poor counts
- inter-technologist differences in ROI drawing
Smoothing methods for curves
- manual smoothing
- moving average
- weighted moving average
Manual smoothing
- simple
- time consuming
- human bias
Moving average
- three point smooth
- ynew = y(n-1) + y(n) + y(n+1)/3
- five point smooth
weighted moving average
- as data moves further from principle point, they are given less weight
- Ynew = 1/∑wi x ∑wiYi
Regression analysis
- a more sophisticated version of smoothing
- analytic smoothing
- assumes curve should follow a mathematical function
- derives new curve according to the desired mathematical function with the least amount of deviation from the original data
- aka least squares curve fitting
Smoothing methods for wall motion studies
- spatial smoothing
- temporal smoothing
- count normalization
- count poor wall motion images are enhanced before the cine is run
- may go through more than one post-processing program