Image Enhancement Biomedical Image Analysis Rangaraj M Rangayyan

Image Enhancement Biomedical Image Analysis Rangaraj M. Rangayyan course: biomedical image processing vibhor kumar Hannu Laaksonen

Topics to be covered 1) Convolution mask Operations 2) . unsharp masking 3) . Sobtracting Laplacian 4) 2) High Frequency Emphasis 5) 3) Homomorphic filtering for Enhancement 6) 4) Adaptive Contrast enhancement

Convolution Mask Operators - Unsharp masking The Generalized equation of unsharp masking is f(m, n) = [g(m, n) - µg(m, n)] + g(m, n) blurred image Is calculated as average of the pixels in the window taken around the pixel(m, n) The weight can be changed according to desired effect. For e. g. For a 3 X 3 convolution mask the unsharp masking is given by - 1 - 1 8 1 8 8 8 1 8 - 2 - - 1 - 8

Subtracting Laplacian The degraded image may be expressed in a Taylor series as 2 2 g g g(x, y, ) = g(x, y, 0) + (x, y, t) + …. . . 2 t Taking g = k 2 g using t diffusion model 2 We get fe = g - k g For k = 1 and mask 3 X 3 we get subtracting Laplacian as: 0 -1 0 Unlike Laplacian subtracting Laplacian -1 5 -1 0 maintain the intensity information while making the image sharp

(a) (b) (a) original lena image (b) Laplacian (c) Unsharp masking (d) subtracting laplacian (c) (d)

High-frequency Emphasis Highpass filtering are useful in detecting edges but for enhancing the images it is necessary to maintain the intensity information. High-emphasis filter does the image enhance keeping the intensity information. The Butterworth high-emphasis filter can termed as: H(u, v) = 1 + 2 1 + (sqrt(2)-1 ) Filter gain Frequency 2 n D 0 D(u, v)

(a) Original shape image (b) the ideal high pass filter (c) The Butter worth highpass filter (d) the Butterworth highemphasis filter

Enhancement using Homomorphic filtering transform input image linear filtering and enhancement inverse transform filtered image

(a) Original Image (b) log transform of original image (c) Homomorphic filtering including a Butterworth high-emphasis filter (d) Butterworth high imphasis filter only (e) Graphical Models and Image Processing, 54(3): 259 -267, May 1992

Adaptive Contrast Enhancement Adaptive-neighborhood contrast enhancement: (1) non overlapping regions segmentation (2) Overlapping regions segmentation seed fill region growing: The region consists of spatially connected pixels that fall within the specified gray level devaiation from seed pixel. every time data is devided into back ground and foreground pixels The growth tolerance threshold is highly important factor

Adaptive Contrast Enhancement contrast enhancment can be done using the formula fe = b (1+Ce)/ (1 -Ce) mean background value increased contrast (a) Part of mammogram with a cluster of calcification (b) adaptive-neighborhood contrast enhancement (c) gamma correction (d) unsharp masking (e) IEEE Transcation on Medical imaging 11(3): 392 -406, 1992

Topics covered 1) Convolution mask Operations 2) . unsharp masking 3) . Sobtracting Laplacian 4) 2) High Frequency Emphasis 5) 3) Homomorphic filtering for Enhancement 6) 4) Adaptive Contrast enhancement