1. Similarity-Based Image Retrieval for Typhoon Images

This site puts emphasis on image search technology as the image database of typhoon imagery. This is based on the similarity of images, and we can search for images based on the cloud patterns of typhoons. The following shows examples of searching similar images by setting the query image randomly.

Western North Pacific (Typhoon)
Western South Pacific (Cyclone)

On the birth of typhoons, the latest typhoon images are shown on the top page, and at the same time they serve as key images, or queries, to find similar images in the past. Then each clicking of an image initiates the similarity-based search of images. This is the result of typhoon image search by our image search engine based on the similarity of cloud patterns between the query image and past typhoon images.

What is the necessity of such functionality? Before explaining you the detail, let us briefly review the state-of-the-art image search techniques. They are roughly classified into two categories as follows:

Searching images using keywords assigned on images or related text. Images can be searched, however, this method is basically similar to text search techniques. Practical services such as Google image search and others are usually based on this kind of techniques.
Searching images without keywords but by means of information that is contained in the image itself such as pattern and color. This technique enables searching images using information hard to be expressed in keywords, such as image pattern.

Typhoon image search is based on the latter technique, or on image information itself such as cloud patterns. This approach is called content-based image retrieval. Then why do we need such a function? The primary reason is that typhoon cloud patterns actually involve precious information for practical uses.

For example, we often hear the central pressure of the typhoon on weather programs, but those values are usually not measured values, but estimated values from cloud patterns. This is natural if you imagine the danger of going into the center of the typhoon in the middle of severe storms just to measure the central pressure. An idea is already suggested to use an autonomous airplane to measure inside the typhoon, but it is still not practical. Hence the only way to estimate the intensity of the typhoon is to use satellite images, observing the entirety of the typhoon from space, and estimate the intensity based on the appearance of the typhoon cloud pattern. This method is called Dvorak method.

This method is based on an assumption that similar-looking typhoons may have similar intensity. Of course, this is a too-simplistic description, and the real procedures are more complex and based on many sources of data. Basically speaking, however, the procedures are based on the experts' analysis and judgment estimating the intensity of the typhoon comparing current cloud patterns with many patterns in the past.

In this sense, an image database with searching functionality directly using cloud patterns may lead to interesting uses and discovery. This is a basic idea of the typhoon image search.