Local Directional Number Pattern for Face Analysis: Face and Expression Recognition

ABS RA! This paper proposes a novel local feature descriptor, local directional number pattern (LDN), for face analysis, i.e., face and expression recognition. LDN encodes the directional information of the faces textures (i.e., the textures structure) in a compact pattern ay, producing a e encode more discriminative code than current methods. !e compute the structure of each micro" ith the aid of a compass mas# that extracts directional information, and such information using the prominent direction indices (directional numbers) and sign$ hich allo s us to distinguish among similar structural patterns that have different intensity transitions. !e divide the face into several regions, and extract the distribution of the LDN features from them. Then, e concatenate these features into a feature vector, and e use it e as a face descriptor. !e perform several experiments in test our descriptor tas#s. hich our descriptor performs

consistently under illumination, noise, expression, and time lapse variations. %oreover,

ith different mas#s to analy&e its performance in different face analysis

AR!"# E! $RE

Existing System 'n this (xisting )ystem, This recognition problem is made difficult by the great variability in head rotation and tilt, lighting intensity and angle, facial expression, aging, etc. )ome other attempts at facial recognition by machine have allo ed for little or no variability in these *uantities. +et the method of correlation (or pattern matching) of unprocessed optical data, hich is often used by some researchers, is certain to fail in cases here the variability is great. 'n particular, the correlation is very lo ith t o different head rotations. Disad%antage !here face recognition does not or# ell include poor lighting, sunglasses, long bet een t o pictures of the same person

hair, or other ob,ects partially covering the sub,ects face, and lo resolution images. -nother

serious disadvantage is that many systems are less effective if facial expressions vary. (ven a big smile can render the system less effective. Proposed System 'n this .roposed )ystem, e propose a face descriptor, Local Directional Number .attern (LDN), for robust face recognition that encodes the structural information and the intensity variations of the faces texture. LDN encodes the structure of a local neighbourhood by analy&ing its directional information. /onse*uently, neighbourhood, in eight different directions directions, descriptor for different textures distinguish intensity changes. Ad%antage 0. 1obust against illumination changes 2. .erformance 3etter 4. /ompact %ode AL&'R# "( ) PR#N!#PAL !'(P'NEN ANAL*S#S ./- finds a linear pro,ection of high dimensional data into a lo er dimensional subspace such as5 (odules 0. Local Direction Number .attern (LDN) 2. :ace (xpression The variance retained is maximi&ed. The least s*uare reconstruction error is minimi&ed. L)'5 Latent )emantic 'ndexing. 6leinberg78its algorithm (compute hubs and authority scores for nodes). 9oogle7.age 1an# algorithm (random al# ith restart). 'mage compression ((igen faces) Data visuali&ation (by pro,ecting the data on 2D). e compute the edge responses in the ith a compass mas#. Then, from all the

e choose the top positive and negative directions to produce a meaningful ith similar structural patterns. This approach allo s us to

a. (igen :aces b. :isher :aces 4. 1esults bet een (igen and :isher :aces (odules Description +, Local Direction Number Pattern 'n this module, LDN is a six bit binary code assigned to each pixel of an input image that represents the structure of the texture and its intensity transitions. e create our pattern by computing the edge response of the neighbourhood using a compass mas#, and by ta#ing the top directional numbers, that is, the most positive and negative directions of those edge responses. -, Face Expression 'n this %odule, each face is represented by a LDN histogram (L8). The L8 contains ;ne to coarse information of an image, such as edges, spots, corners and other local texture features. 9iven that the histogram only encodes the occurrence of certain micro"patterns ithout location information, to aggregate the location information to the descriptor. a, Eigen Faces 'n this module, facial recognition is discriminating input signals (image data) into several classes (persons). The input signals are highly noisy (e.g. the noise is caused by differing lighting conditions, pose etc.), yet the input images are not completely random and in spite of their differences there are patterns hich occur in any input signal. )uch patterns, ell as relative distances hich can be observed in all signals could be " in the domain of facial recognition " the presence of some ob,ects (eyes, nose, mouth) in any face as bet een these ob,ects. These characteristic features are called eigenfaces in the facial recognition domain (or principal components generally). They can be extracted out of original image data by means of a mathematical tool called .rincipal /omponent -nalysis (./-). 3y means of ./- one can transform each original image of the training set into a corresponding eigenfaces. -n important feature of ./- is that one can reconstruct any original image from the training set by combining the eigenfaces. 1emember that eigenfaces are nothing less than characteristic features of the faces. Therefore one could say that the

original face image can be reconstructed from eigenfaces if one adds up all the eigenfaces (features) in the right proportion. b, Fis.er Faces 'n this %odule, bit harder to explain, because they identify regions of a face that separate faces best from each other. None of them seems to encode particular light settings< at least it=s not as obvious as in the (igenfaces method. 'f ' could only guess describes hich features> )o the :isher faces method for sure, is the ability to reconstruct faces. 'f ' faces, ,ust li#e in the (igenfaces section. /, Results bet0een Eigen and Fis.er Faces /omputational complexity (ffectiveness across pose )ensitivity lighting :isher faces )lightly complex 9ood, even limited data to Little (igen :aces more )imple ith )ome data very ith enough hich component e lose ith ant to reconstruct e leave the interpretation up to the reader. !hat

"ARD1ARE RE2$#RE(EN S )ystem 8ard Dis# %onitor 5 5 5 .entium '? 2.@ 98&. AB 93. 0C ?9- /olour.

%ouse 1am

5 5

Logitech. C02 %3.

S'F 1ARE RE2$#RE(EN S Dperating system :ront (nd /oding Language 5 5 5 !indo s A (42"3it) ?isual )tudio 2B0B /E.N(T