Colors---Messengers of Concepts: Visual Design Mining for Learning Color Semantics

Abstract

Colors are cognitive: Perceived by our visual system, colors are classified at higher levels of abstraction into verbal and semantic categories. In practice, for instance in visual design, color combinations are designed not only to be appealing, but also to be silent salespersons that communicate with the audience. In this talk, I discuss my work on the concept of color semantics.

I first introduce a collection of magazine covers, as professional design examples, in variety of different genres. I then describe an extension to the LDA topic modeling framework to express semantic concepts as a combination of color and word distributions -- color-word topics. While LDA models text documents as distributions over word topics, we model magazine covers as distributions over color-word topics. To evaluate the strength of the model, I describe the design of two crowdsourced experiments for matching colors with words and vice versa. Finally, I demonstrate several simple prototypes that apply the learned model to color palette recommendation, design example retrieval, and image recoloring.

Bio: Ali Jahanian is a Postdoctoral Associate at CSAIL MIT, working with Prof. Fox Harrell on understanding aesthetics of virtual identities across platforms. Recently, Ali did a postdoc with Dr. Ruth Rosenholtz in the Vision and Graphics Area of CSAIL, focusing on applications of human vision in visual design. Prior to MIT, Ali did a PhD in the School of Electrical and Computer Engineering at Purdue University, under the supervision of Prof. Jan P. Allebach. Ali's PhD thesis focused on the automatic design of visual media as well as the quantification of such designs' aesthetics. Ali's main interest is in developing vision models to learn aesthetics of visual design, and tools to help non-designers in visual design of media for self-publishing. His work proposes approaches towards escalating creativity and inspiration for non-designers by providing design alternatives and recommendations.