mip1_05_image_enhancement_090519_1462589

When working with 8 bit digital images the pixel

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: brated intensity space”), where the data is stored. When working with 8-bit digital images the pixel values range is [0 255] ([0 amax]) rather than [0 1], thus we additionally have to perform proper scaling: 19.05.2009 Dr. Pierre Elbischger - MIP1/ISAP'SS09 28 Automatic Contrast Adjustment Contrast enhancement is achieved by mapping the current darkest and brightest pixel value to the lowest and highest possible. alow and ahigh are the lowest and highest pixel value in the current image, whose full intensity range is [amin amax]. In order to make the algorithm more robust to outliers (extremely high/low pixel values) a fixed percentage (slow, shigh) of pixels is saturated to the upper and lower ends of the target intensity range. original auto-contrast Matlab: imadjust 19.05.2009 Dr. Pierre Elbischger - MIP1/ISAP'SS09 29 Histogram Equalization (1) Histogram equalization is a method for automatic contrast improvement. Idea: Improve the contrast by applying a monotonic gray-level transform T(.) to an input image with histogram H(r) that results in an output image with uniformly distributed Histogram G(r), thus they use the entire dynamic range. G(r) H(r) r r ? For our considerations the histogram can be treated as a continues function (similar to a PDF). The monotonically increasing property of the transform implies: 19.05.2009 Dr. Pierre Elbischger - MIP1/ISAP'SS09 30 Histogram Equalization (2) Assuming an image of size NxN with the input brightness range p0 to pu, the equalized PDF G(r) corresponds to the uniform PDF with constant value over the entire output brightness range q0 to qu (all N2 input pixels are equally distributed of the entire output gray-scale range. Substitution into the first equation results in The desired transformation can than be derived by solving for q: The integral in the transform is the distribution function an can be replaced by the cumulative histogram for digital images: Matlab: histeq 19.05.2009 Dr. Pierre Elbischger - MIP1/ISAP'SS09 31 Hi...
View Full Document

This note was uploaded on 07/09/2009 for the course MEDIT 1 taught by Professor Pierreelschbinger during the Spring '09 term at Carinthia University of Applied Sciences.

Ask a homework question - tutors are online