Two very important problems in setting up an efficient digital library system containing thousands of high-resolution images are efficient storage and fast retrieval. In this context pre-processing of the image data is important, and and we will address issues related to storage, indexing for retrieval, and fast browsing of low resolution data. The last step can be viewed as search by the user over a smaller dataset after the catalogue section narrows down the domain to a reasonable size.
A typical black-and-white image of size 512x512 pixels with a 8 bit resolution contain over 2 million bits. The storage problem gets much worse for multi-spectral images of larger size, and for color images. As a result, it is critical to develop efficient storage systems for a digital image library. Browsing through a very large number of such images is also very time-consuming before the user can decide which particular image is of interest for further study and use. One classical approach, followed in many commercial image processing systems, is to store separately highly decimated versions of the original images. As these images are of smaller sizes, a number of them can be displayed simultaneously for browsing. For example, if the original image size is of 1024x1024, each such image can be decimated by a factor of 32 in both horizontal and vertical directions resulting in a low-resolution image of size 32x32. If the display is of size 1024x1024, 32 such low resolution images can be displayed together at one time. However, such an approach is not memory- efficient as both the high resolution original and the low-resolution images are being stored.
In this project we propose to solve both of these problems by employing a wavelet decomposition approach [7]. This permits an iterative decomposition of an image into a number of subimages of progressively lower resolutions, and images of higher resolutions can be formed by selectively combining these lower resolution sub-images. The generation of the higher resolution images must be done through the appropriate synthesis filter bank to guarantee perfect reconstruction of the original image. An advantage of the wavelet decomposition is that the storing of all sub-images require the same amount of memory as the original image and hence this approach is memory-efficient. Another advantage of this approach can be achieved by storing the subimages in progressively slower memories as the resolution increases. For example, the lowest resolution image can be stored in a very high speed memory for fast access.