A novel approach is proposed for quick and accurate partial image retrieval from a large number of images based on tree structure of images. Retrieving the partial similar images generally requires a huge amount of storage space for indexes due to the large number of portions of images. The proposed method constructs the ternary tree, which greatly reduces the storage space for database images and fast retrieval. The computation of tree structure is not a computationally intensive task and tree comparison won't take much time to find the best match image. The system is capable of accepting any size of query image decided by user and retrieves the whole images from the database. The retrieval time is of the order of seconds ranging from 42 milliseconds to 10 seconds and translation independent.