Abstract: Generating a color theme according to users’text input is one way to automatically obtain the color theme and is important to image processing and art design.A method of text-driven color palette generation using Internet images is proposed,based on the characteristics of the problem and the shortcomings of existing approaches.The proposed method consists of 3 steps.First,it collects many images from the Internet or image database using users’text input as a search keyword and then generates a color palette for each collected image depicting its color distribution.Second,it organizes the color palettes for all collected images into clusters with the affinity propagation clustering method,then sorts them according to the cross-group difference index proposed in this paper.Third,it picks one representative palette for each cluster among a few top clusters,forming the final color palette candidates result.The whole method is simple,easy to use,and able to generate semantically diverse color themes.The user study results show that compared with the method called Text2Colors,the color themes generated by this paper’s method are more semantically correct and more diverse.For 69%and 94%of the input text in the randomly selected input text data set,more than half of the users think that this paper’s method has better accuracy and diversity,respectively.