楊鵬穎

This paper proposes a method that can be used for image emotion identification which uses colors and human facial expression as features since these features can directly influence observers. Color feature is based on Color Image Scale, which is proposed by Kobayashi. Color Image Scale is a system that uses 130 basic colors to make different three-color combinations and there are high-level semantic concepts for each combination. These semantic concepts contain “modern”, “classic”, “natural”, etc. Moreover, these concepts are in a 2-dimensional space and grouped by perceived similarity. Facial expression feature is based on Ekman’s research, which proposed six basic emotions include “happiness”, “sadness”, “fear”, “anger”, “surprise”, and “disgust”. The six basic emotions are cross-cultural so that this paper will not consider influence of different races of people in the image. However, even though the basic emotion can be used in general model, there still exist differences for people to express their emotions. Therefore, this paper uses geometric feature, which is more general than appearance feature, to classify the facial expressions. After extracting color and facial expression feature, SOM (Self-Organized Map) is used for classifying the image emotion with the two features. The intended applications are image realization and information extraction.