Statistical Methods for Reducing Uncertainty and Eliminating Distortion in Type2 Fuzzy Sets
Reshma Kar, Mandakranta Chakraborty, Diptendu Bhattacharya · 2012
Abstract:. A fuzzy scheme of emotion recognition seems intuitively suited to the fuzziness present in human emotions. Type 2 Fuzzy sets have been effectively employed to recognize emotions of a person. The effectiveness of Interval Type2 Fuzzy Sets has been studied on the Japanese database. By employing statistical techniques like non uniform sampling and modal approach, computational complexity was reduced while accuracy of the algorithm was increased. The non uniform sampling technique improved accuracy by eliminating distorted fuzzy memberships obtained due to interpolation. The computational complexity was reduced by plotting fewer points on the primary Gaussian curve. Further only the modal values within a range of the membership interval were considered. This technique further reduces the uncertainty and hence improves accuracy rate. I. INTRODUCTION Human comprehension and understanding is deeply co-related with emotions. For example failure is accompanied by sadness, and success accompanied by happiness; the act of perceiving by humans may be disturbed, when they suffer from fear, and so on. Hence the interaction between human beings and computers will be more natural if computers are able to perceive and respond to human emotions. With the help of emotional intelligence many human -computer issues can be resolved. The paper provides two complementary approaches to improve emotion recognition from an unknown facial expression, when the emotion class of individual facial expression of a large number of experimental subjects is available. This paper introduces the concept of non-uniform sampling to enhance the accuracy of the method. In particular, General Type2 Fuzzy Sets (GT2FS) and Interval Type2 Fuzzy Sets (IT2FS) have been studied. Section II introduces non- uniform sampling technique improves the accuracy of the previous algorithm at the same time reduces the computational and tine complexity by a large amount. Section III introduces and explains the implementation of modal range method on Interval Type-2 fuzzy set (IT2FS) which shows an improvement in accuracy of approximately 5% on both IT2FS and .Section IV gives the conclusion. II. ELIMINATION OF DISTORTION IN FUZZY MEMBERSHIP VALUES To obtain a smooth curve as in Fig 2(a), between the intervals 0.1 and 0.01, a sampling interval of 0.0001 has to be taken. This means that the number of primary memberships obtained will be 901, for which number of comparisons in DE will be high. For this reason, running time of this algorithm is very high, approximately 1 hour for 5 distributions, which has to be repeated for 6 features and 6 emotions, totaling to approximately 36 hours. If the sampling interval is reduced, to reduce the time consumption, the graphs become distorted as shown in Fig 2(b), Hence a nonuniform sampling method is proposed, which retains the basic shape of the graph.