Outline • Announcement • Texture modeling - continued – Some remarks – Applications of texture modeling Announcement • The presentation schedule is on the web – Now you should have almost completed your project – You need to take it very seriously in order to get a good grade for this class 5/29/2016 Visual Perception Modeling 2 Comments on General Feature Statistics 5/29/2016 Visual Perception Modeling 3 Joint Statistics • FRAME and Julesz ensemble models use marginal distributions of feature statistics • It might be useful to consider joint statistics for more powerful models – Joint statistics will be more precise because filter responses are not independent of each other – However, this model should include all the images of the same texture type; an overconstrained model will include only the original image 5/29/2016 Visual Perception Modeling 4 Multi-resolution Sampling 5/29/2016 Visual Perception Modeling 5 Multi-resolution Sampling – cont. 5/29/2016 Visual Perception Modeling 6 Multi-resolution Sampling – cont. More results at http://www.ai.mit.edu/~jsd 5/29/2016 Visual Perception Modeling 7 Applications of Texture Models • Inspection – There has been a limited number of texture processing for automated inspection problems – Detection of defects of textiles – Detection of defects of lumber wood automatically 5/29/2016 Visual Perception Modeling 8 Applications of Texture Models – cont. • Medical image analysis – Image analysis techniques have played an important role in several medical applications – Texture features are used to distinguish normal tissues from abnormal tissues 5/29/2016 Visual Perception Modeling 9 Applications of Texture Models – cont. 5/29/2016 Visual Perception Modeling 10 Applications of Texture Models – cont. • Document processing – Document image analysis and character recognition • Applications ranging from postal address recognition to interpretation of maps – Based on the characteristics of printed documents 5/29/2016 Visual Perception Modeling 11 Applications of Texture Models – cont. • Remote sensing – Texture analysis has been used extensively to classify remotely sensed images • Land use classification • Automated image analysis 5/29/2016 Visual Perception Modeling 12 Applications of Texture Models – cont. 5/29/2016 Visual Perception Modeling 13 Applications of Texture Models – cont. • Content-based image retrieval – Try to retrieve images that are meaningful in certain sense • For example, to find all the images that like the examples • To find all the images that contain a horse 5/29/2016 Visual Perception Modeling 14 Applications of Texture Models – cont. 5/29/2016 Visual Perception Modeling 15 Content-based Image Retrieval • Image retrieval example using spectral histogram http://www-dbv.cs.uni-bonn.de/image/mixture.tar.gz 1st (Distance: 0.05) 5/29/2016 6th (Distance: 0.14) Visual Perception Modeling 12th (Distance: 0.21) 16 Applications of Texture Models – cont. • Texture segmentation – Image segmentation is to partition an image into roughly homogenous regions – Segmentation is more difficult than classification • Feature statistics not known • Boundaries to be localized 5/29/2016 Visual Perception Modeling 17 Texture Segmentation - continued • Identify feature statistics using spatial constraints – Pixels within a homogenous region have similar spectral histogram 5/29/2016 Input image Visual Perception Modeling Initial regions 18 Texture Segmentation - continued • Classify each pixel using the extracted feature statistics Initial classification result Error from the ground truth – Error with respect to the ground truth is 6.55 % 5/29/2016 Visual Perception Modeling 19 Texture Segmentation - continued • Boundary localization using structural information Segmentation result Error from the ground truth – The segmentation error is 0.95 % 5/29/2016 Visual Perception Modeling 20 Texture Segmentation - continued 5/29/2016 Visual Perception Modeling 21 Texture Segmentation - continued 5/29/2016 Input image Result superimposed Canny edge map Segmentation result Visual Perception Modeling 22