Outline • Neural networks - reviewed – Back-propagation program • Texture modeling – Introduction Back Propagation Program • Programs – Backprop.c – main program – Propagation.c – contains procedures for BP – Para-util.h and type-def.h – contain data structure definitions – Located at ~liux/public_html/courses/research/programs/neural-networks • Parameter files – Control parameter file – network-3-3-1.par – Training data file – network-3-3-1-training.par 5/29/2016 Visual Perception Modeling 2 Back Propagation Program – cont. • Homework #5 – Gain some first-hand experience with neural networks – Study how the parameters affect the performance of neural networks 5/29/2016 Visual Perception Modeling 3 Texture Modeling • Texture is a phenomenon – Is widespread – Easy to recognize – Hard to define as many other perceptual phenomena • Texture arises from different resources – Views of large numbers of small objects • Grass, brush, pebbles, hair, ...... – Surfaces with orderly patterns • Cheetah skins, zebra stripes, ...... 5/29/2016 Visual Perception Modeling 4 Some Texture Examples 5/29/2016 Visual Perception Modeling 5 Non-texture Examples 5/29/2016 Visual Perception Modeling 6 Texture Definition • Image texture is defined as a function the spatial variation in pixel intensities – Local statistics or local properties are constant, slowly varying, or approximately periodic 5/29/2016 Visual Perception Modeling 7 Deterministic textures • Deterministic textures – A set of primitives – A placement rule – Examples include • A tile of floor • Regular structures 5/29/2016 Visual Perception Modeling 8 Stochastic Textures • Stochastic textures – Do not have easily identifiable primitives – However, there are local statistics/local properties that are varying slowly or approximately periodic 5/29/2016 Visual Perception Modeling 9 Texture Modeling • Texture modeling is to find feature statistics that characterize perceptual appearance of textures • There are two major computational issues – What kinds of feature statistics shall we use? – How to verify the sufficiency or goodness of chosen feature statistics? 5/29/2016 Visual Perception Modeling 10 Texture Modeling – cont. • The structures of images – The structures in images are due to the inter-pixel relationships – The key issue is how to characterize the relationships 5/29/2016 Visual Perception Modeling 11 Psychophysical Texture Models • Texture discrimination 5/29/2016 Visual Perception Modeling 12 Psychophysical Texture Models – cont. • Julesz conjecture – Two textures that have identical second-order statistics are not pre-attentively discriminable • Second-order statistics – First-order statistics are the histogram of the texture images – Second-order statistics are defined as the likelihood of observing a pair of gray values occurring at the endpoints of a dipole 5/29/2016 Visual Perception Modeling 13 Co-occurrence Matrices • Gray-level co-occurrence matrix – One of the early texture models – Was widely used – Suppose that there are G different gray values in a texture image I – For a given displacement vector (dx, dy), the entry (i, j) of the co-occurrence matrix Pd is Pd (i, j ) | {( r , s) : I (r , s) i, I (r dx, s dy) j} | 5/29/2016 Visual Perception Modeling 14 Co-occurrence Matrices – cont. • Properties – Size of the co-occurrence matrix is G x G – The co-occurrence matrix in general is not symmetric • A symmetric version can be computed as Pd P d Pd – The co-occurrence matrix reveals certain properties about spatial distribution of the gray levels in the texture images 5/29/2016 Visual Perception Modeling 15 Co-occurrence Matrices – cont. • Useful texture features – Because the co-occurrence matrices can contain many entries, a number of features are proposed to calculate from co-occurrence matrices • Energy 2 P d (i, j ) i • Entropy i Pd (i, j ) log Pd (i, j ) i • Contrast j 2 ( i j ) Pd (i, j ) i 5/29/2016 j Visual Perception Modeling 16 Co-occurrence Matrices – cont. • Generalization of co-occurrence – k-gon statistics – In general, we can define an arbitrary polygon with k vertices and collect statistics on those vertices • A line segment defines the co-occurrence • A triangle defines 3-gon statistics – It captures the dependence among pixels 5/29/2016 Visual Perception Modeling 17 Autocorrelation Features • Autocorrelation features – Many textures have repetitive nature of texture elements – The autocorrelation function can be used to assess the amount of regularity as well as the fineness/coarseness of the texture present in the N N image I (u, v) I (u x, v y ) ( x, y ) u 1 v 1 N N 2 I ( u , v ) u 1 v 1 5/29/2016 Visual Perception Modeling 18 Geometrical Models • Geometrical models – Applies to textures with texture elements – Then one can compute the statistics of local elements or extract the placement rule that describes the texture – Voronoi tessellation features – Structural methods 5/29/2016 Visual Perception Modeling 19