Segmentation Using Texture 1 Project Description l Input

Segmentation Using Texture 1

Project Description l Input: satellite image and a texture l Task: segmentation of the image based on the texture l Output: labeled image 2

What Is a Texture ? There are many definitions of the word texture: v Describes something that has a surface that is not smooth but has a raised pattern on it (from Cambridge advanced learner's dictionary) v A measure of the variation of the intensity of a surface, quantifying properties such as smoothness, coarseness and regularity (from FOLDOC - computing dictionary) l 3

Algorithms l Histogram matching l Law’s texture measure l Run-length matrices 4

Histogram Matching Algorithm I Short description: Window at step k (the sample) The texture we are searching (the template) Window at step k+1 The basic idea is to compute the histogram of the template, and then sweep a window over the image, compute the histogram of the window and do a correlation between the histograms. 5

Histogram Matching Algorithm II i. Histogram equalization (HE) of the image: Calculate the histogram of the texture iii. Overlap the image by the texture at each possible position and calculate correlation of the histogram of the texture f and the one of the overlapped area g: ii. FOR MORE INFO. . . Histogram Transformation in Image Processing and Its Applications by Attila Kuba, University of Szeged 6

Histogram Matching Algorithm III iv. Thresholding of the correlation map: i. High correlated values are set to 1 ii. Low correlated values are set to 0 This yields a binary image BI v. Median filter to eliminate the holes on BI vi. Border : = BI – erosion(BI) vii. Put the border on the original image OBSERVATION. . . You can choose an algorithm for the search (we have more than one ) You should wait (but not too long) for the resulting image 7

Histogram Matching Algorithm IV Zoomed texture 8

Histogram Matching Algorithm V Zoomed texture 9

Run-length Algorithm I City – rough grayscale variations – short runs =P Grass – smooth grayscale variations – long runs =P 10

Run-length Algorithm II Second step: ü Calculate short run emphasis ü Calculate long run emphasis ü Calculate gray level nonuniformity ü Find closest matches FOR MORE INFO. . . Tang, Xiaoou, “Texture Information in Run-Length Matrices”, IEEE transactions on image processing, vol. 7, no 11, november 1998 http: //www. s 2. chalmers. se/undergraduate/courses 0203/ess 060/PDFdocuments/For. Printer/ 11 Notes/Texture. Analysis. pdf

Law’s Texture Measure I First step: Vertical kernel Measure energy Horizontal kernel Measure energy Original image Law’s energy matrix FOR MORE INFO. . . Chantler, Michael J, “The effect of variation in illuminant direction on texture classification”, pp 90 -134, http: //www. cee. hw. ac. uk/~mjc/texture/mjc-phd/ 12

Law’s Texture Measure II Second step: Grayscale dilation Thresholding Binary dilation Law’s energy matrix Segmented image FOR MORE INFO. . . Krabbe, Susanne, “Still Image Segmentation”, http: //wwwmm. informatik. unimannheim. de/veranstaltungen/animation/multimedia/segmentation/documentation/Seg 13 mentation. pdf

Law’s Texture Measure III Original image Output image 14

Blaž Luin Kornél Kovács Dumitru Şipoş Zoltán Kiss 15