Studies on Classification of Similar Trademarks Corresponding to Human Visual Cognitive Sense

摘要

In this paper we are concerned with classification of similar trademarks corresponding to human visual cognitive sense. When similar trademarks are classified or retrieved using computer, physical features are extracted from images of trademarks, physical feature space is constructed based upon the extracted physical features and similarity between trademarks is evaluated by the distance in the physical feature space. The results of retrieval or classification however do not always coincide with human visual cognitive sense of similarity. We have been therefore studying methods to reflect human visual mentality in retrieval of similar trademarks. In these methods a questionnaire is set out to obtain the data on human evaluation of similarity between trademarks, the physical feature space is transformed to the subjective feature space which reflects human visual mentality using the data on human evaluation obtained by questionnaire and similar trademarks are retrieved in the subjective feature space. The efficiency of this transformation has been verified but here in order to consider a method to reflect human visual mentality in classification of similar trademarks without questionnaires a function is used to express human visual cognitive sense, which is called "autocorrelation function". Since it is considered that human judgement on similarity is influenced by the pattern of trademarks and the autocorrelation function is able to express the pattern of trademarks, it is studied how to classify trademarks according to human visual cognitive sense of similarity by using the autocorrelation function.