[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
In this paper, we propose a method to use several images, which are synthesized from images in a database, as examples for communicating design imagery to image designers from their clients to make teaching materials and so on. We suppose that these imageries correspond to various kinds of image features and it is difficult to represent them by a single image.
Recently, with the progress of multimedia technology, we become to use visual contents frequently. Since most of those who want to make such visual contents do not have the skill of painting and drawing images, they have to ask designers to make the contents that they want. In those cases, the designers usually communicate with their clients by linguistic expression in order to know their design imageries of them. However, due to the difference in expressions between such linguistic expressions and images, the designers have to make trial and error by making image samples through the communication with their clients. This process becomes troblesome especially when the designers need a lot of time to make each sample images.
It would improve the efficiency of this process if images can be used for the communication between designers and clients because the same expression is used for the communication and image contents. But a single image as an example of the client's imagery is not sufficient to communicate the imagery.
In order to get several images as examples of the client's imagery, we suppose an images space, which is a space with the axes corresponding to the feature parameters of images. Clients and designers share the same image database. Images in the database are scattered around the space, and the design imagery of a client may correspond to a region in this space. The images as examples to communicate the imagery are obtained by sampling this region.
In this sampling, first, we must find this region from the image space because clients usually do not recognize the border of the region clearly. Then, we have to select some images among those in that region. In order to transmit the region properly, those selected images should not cluster together. And in some cases, those images may not exist in the database. Thus, the images need to be synthesized as combinations of image regions, which are called object, of the images in the database.
In this paper, we construct the system which can realize these processes mentioned above. The design imagery of image content varies in various levels, from an abstract level such as overall impressions of the image, to a concrete level mutually related to the image features including color, shape, and layout. We focus on the concrete level of imagery, and aim to realize the above processes based on the color features.
In our system, the client draws a rough sketch of his/her imagery, which is called query in this article. The query is composed of several regions, which are called component. The system searches for the objects close to each components in the image database and presents them to the client. The client decides whether each object is acceptable or not. As the result of this process the system acquires a set of objects to constitute example images.
Then the system synthesizes the images by combination of objects selected in the process above and shows them to the client. The client decides whether images are acceptable or not as an example of the client's imagery. Based on the decision of the client, the system determines which combination of objects should be presented to the client next. By repeating this process until several images are obtained so that they are accepted as examples of the client's imagery and are scattered sparsely in the region corresponding to that imagery.
We made experiments to evaluate our method by communicating the imagery dependent mainly on the colors of image. In the experiments, each subject communicated his/her design imagery to another subject with the images created by this system. From the result of these experiments, it is shown that the rate of agreement for the decisions of the two subjects on whether each of the test images is permitted as an example of the imagery is improved compared for the case without communication.