INTEGRATING LOW-LEVEL FEATURES COMPUfAnON WITH INDUCI1VE LEARNING TECHNIQUES FOR TEXTURE RECOGNITION PETER

This paper presents a method for applYIng mducllve learnmg techniques [0 tel{ture description Jnd recogmtlon. Local features of texture are computed by tWO well-known methods. Laws' masks and cO-Q<;currence matnces. Then. a three-level generalizaliOll of local features IS applied (0 .:reate teltture descnpllon rules. The fU"St level genera/ization. the scaling intertace. has been implemented to transform the numenc data of local texture features into their higher symbolic representation as numencal ranges, ThIs scaling interface tests data consistency as well. The ,realton of descnption rules incorporaung the inductlve incremental learning algonthm IS the second generalization step. The SG-TRUNC method of rule reduction is applied as the next hlerarchu:al generalizallon level. ThIs machine learnmg approach to texture description and recognitIon IS compared WIth tile "asSIC pattern recognition methodology, The results from the recogmtlon phase are presented from SIX classes of textures. charactenzed by smoothly ..:hangmg ,llummalJon andlor texture resolution, The average recogmllon rate was 91 °'0 for the mductlve learning approach. and all classes of textures were recognized, In comparison. the tradillonal k;·NN panem recogRition method jid not recogRize one class of texture. and the average recognition rate was 83%. The proposed methodology smooths tile recognition rates through the hierarchy of generalization levels. i.e. the next genera/izallon step increases these rates for classes that were less easdy recognized. and decreases these rates for classes that were more ;:asily recognized.