Semantic Concept Extraction For Eyebrow Shapes Via AFS Clustering

Axiomatic Fuzzy Sets(AFS)theory was proposed first as a new analytic method of fuzzy mathematics.It discussed fuzzy sets and fuzzy logic from a more general and level,which axiomatic the idea of fuzzy sets,and abstracts the concept of human perception and the mechanism of its logical relationship.In the framework of AFS theory,which provides an effective tool to concert the information in the training examples and databases into the membership functions and their fuzzy logic operations.Make the established membership functions and fuzzy logic more objective and unified.At present,AFS theory has been applied to knowledge representation,clustering analysis,fuzzy classification and so on.Aiming at the problem of semantic description and clustering of eyebrows,based on the AFS theory,the eyebrow semantic descriptor is extracted from three parts:extracts the salient features of eyebrows,defines simple concepts and clusters.With an aim to extract the significant characteristics of the eyebrow,solve the problem of describing the semantics of the eyebrows and the unclear boundaries between different cluster.Based on this paper,the following research was done.(1)Aiming at how to describe eyebrow more accurately and extract the fuzzy concepts,the fuzzy clustering algorithm is used to research the semantic descriptor based on AFS theory framework,the main task is clustering and extracting semantic concepts of eyebrow.The specific method is as follows:Firstly,the landmarks of the eyebrows in the face image are extracted.In this paper,the facial landmark detection model is used to automatically extract the feature points of the normalized face image,such as eyebrows and eyes.Secondly,the clustering algorithm based on AFS theory to cluster the landmarks of the detected eyebrows and give semantic concepts.Finally,semantic descriptions of eyebrows are extracted by assigning semantic labels to each eyebrow.The efficacy of this framework is demonstrated on two face databases of BU-4DFE and Multi-PIE databases.At the same time,the effects of different facial expressions and similar facial expressions on the extracted eyebrow semantics were compared when using AFS to extract eyebrow semantics.The experimental results illustrate that the semantic eyebrow descriptors obtained by AFS clustering algorithm are much better than those obtained by traditional clustering methods(k-means and fuzzy c-means)in terms of consistency and comprehension,and they are much closer to human perception.The experimental results show that different facial expressions can affect the extraction of eyebrow semantics,that is,the eyebrows extracted from the same face image have different semantics under different facial expressions.(2)Aiming at how to more accurately represent the shape of the eyebrows and extract the semantic concept of the eyebrow shape.An improved Directional Triangle Area Representation(DTAR)is proposed in this paper.Compared with DTAR,the revised DTAR can select the number of landmarks representing the eyebrows by threshold judgment.The DTAR value of the landmarks of the eyebrows extracted from the facial landmarks detection model is calculated by the improved DTAR,then all the DTAR values representing the eyebrows are connected in an orderly manner to generate a DTAR curve representing the eyebrows.Then,according to how to express the selected reference eyebrows,this paper formulates the method of labeling the reference eyebrows according to the punctuation rules of the facial landmarks detection model and obtains the DTAR curve of the reference eyebrows according to the marked feature points according to the same method.The shape of each reference eyebrow is assigned a semantic label.Finally,the membership functions base on AFS are introduced,which are utilized to measure the degrees of each eyebrow shape similar to the given reference eyebrow,then we extract the semantics by means of combining descriptions which have the largest and the second largest membership degrees.In order to verify the validity of the proposed method and the consistency with human perception,extensive experiments on the AR face database and BJUT eyebrow database are designed and conducted.The experimental results demonstrate that the landmarks of the eyebrows based on the improved DTAR selection are more in line with human perception than direct use DTAR.