PicSOM is a content-based image retrieval (CBIR) system based on the Self-Organizing Map (SOM). It uses a hierarchical version of the SOM called Tree Structured Self-Organizing Map (TS-SOM) as the method for retrieving images similar to a given set of reference images. The system may use one or several types of statistical features for image querying. Separate feature vectors can thus be formed for describing the color, texture, and structure of the images. A separate TS-SOM is then constructed for each feature vector set and these maps are used in parallel to select the best-scoring images.

The image queries are performed through the World Wide Web and the queries are iteratively refined as the system exposes more images to the user. During the process, PicSOM tries to adapt to the user's preferences regarding the similarity of images.

PicSOM online

The demonstration version of PicSOM retrieval system.

Screenshots: [1], [2], [3]

Databases & features

• COIL-100 object database: Random - Color Layout - Edge Histogram - Map hierarchy
• COREL database: Color - User interaction - Segmentation example
• FERET face database: Eigenfaces
• Paper defects: Shape features - Example defect - More info
• Metal defects
• Satellite images
• Multimedia messages (MMS): Example message
• Video retrieval: Multimodal indexing