Digital Photo Album Management Techniques: From One Dimension to Multi-Dimension
Digital Photo Album Management Techniques: From One Dimension to Multi-Dimension
Yang Lu

With the increasing popularity of digital camera, organizing and managing large collection of digital photos effectively are therefore required. In this thesis, we study the techniques of photo album sorting, clustering and compression techniques based on the JPEG DCT frequency domain features, which offers low-cost processing efficiency and excellent texture information.

We utilize the first several non-zero DCT coefficients to build our feature set and calculate the energy histograms. Based on them, we perform image similarity analysis in frequency domain directly without having to decompress JPEG photos into spatial domain first.

We first exploit one dimensional photo album sorting and adaptive clustering algorithms to group the most similar photos one by one. We further compress those clustered photos by a MPEG-like algorithm with variable IBP frames and adaptive search windows. Our methods provide a compact and reasonable format for people to store and transmit their large number of digital photos at the minimal expense of original image quality.

We further study the complex high-dimensional photo album clustering algorithms. We propose to utilize multidimensional scaling (MDS) techniques to solve the high dimensional feature space and unknown number of natural categories problems in more complicated clustering process. We calculate the similarity distances of all pairs of images, then extract the most principal coordinates that reveal how they are related with each other maximally in the compact and observable low dimensional space. Multidimensional Scaling not only provides a favorable visualization layout of all images in terms of the semantic similarity metric adopted, but also suggests the clustering results, which group the most similar photos together visually. With the most significant coordinates generated by MDS, our interactive clustering algorithms have been proved more effective for digital photo clustering and navigation than any traditional clustering algorithms.

Related publication:

  • "Digital Photo Similarity Analysis in Frequency Domain and Photo Album Compression",
    Y. Lu, T. T. Wong and P. A. Heng,
    in Proceedings of The Third International Conference on Mobile and Ubiquitous Multimedia (MUM2004), College Park, Maryland, USA, October 2004, pp. 237-244.

  • "Photo Sorting and Compression in Frequency Domain",
    Y. Lu, T. T. Wong and P. A. Heng,
    in Proceedings of International Symposium on Optics East, Philadelphia, PA, USA, October 2004, pp. 310-319.

Copyright 1996-2012 Tien-Tsin Wong. All rights reserved.