AN IMPRESSION EVALUATION MODEL FOR APPAREL PRODUCT RETRIVAL BASED ON IMAGE ANALYSIS

Abstract

In this paper, we propose an impression evaluation model that can be used for retrieval of apparel data. Using impression words which evoke specific impressions and are reproducible in successive evaluations, we developed a model capable of estimating impression evaluation values from the physical features of apparel images extracted through image analysis. These values function as initial retrieval data. Employing these impression words, we used the Semantic Differential method to analyze subjects' impressions of apparel images. We confirmed the reproducibility of ten pairs of impression words such as “childish-mature”by correlating the results of two impression evaluations by the same subjects conducted one year apart. Spatial frequency analysis and color image processing were used to analyze the color space data of apparel images. From the results, we obtained three types of features: silhouette, hue and pattern. The coefficients for models combining different features were obtained through regression analysis to generate impression evaluation models. Evaluation values for six pairs of impression words exhibited a correlation of 0.7 or more with the values estimated by model F. Of these, the “feminine-mannish”pair was highly influenced by thesilhouette feature while five other word pairs including “showy-plain”were more influenced by the pattern feature.