Learning Graphs to Model Visual Objects Across Different Depictive Styles

In this paper we model visual classes using a graph with multiple labels on each node; weights on arcs and nodes indicate relative importance (salience) to the object description. Visual class models can be learned from examples from a database that contains photographs, drawings, paintings etc. Experiments show that our representation is able to improve upon Deformable Part Models for detection and Bag of Words models for classification.

Fig. Our multi-labeled graph model with learned discriminative weights, and detections for both photos and artworks. The model graph nodes are multi-labeled by attributes learned from different depiction styles (feature patches behind the nodes in the figure). The learned weight vector encodes the importance of the nodes and edges. In the figure, bigger circles represent stronger nodes, and darker lines denote stronger edges. And the same color of the nodes indicates the matched parts.

Learn More...

Modelling Visual Objects Invariant to Dipictive Styles

In this paper, we investigate a method for modelling visual objects classes in a manner that is invariant to depictive style. The assumption we make is that an object class is characterised by the qualitative shape of object parts and their structural arrangement. Hence we use a graph of nodes and arcs in which qualitative shapes such as triangle, square, and circle to label the nodes.

Fig. Examples of three graph models generated from 3 categories of objects, which are horses, bicycles, and butterflies. The visualization shows of selected levels below the corresponding model, with the simple shapes fitted. Child-parent arcs are in blue, adjacencies between the nodes in the same level are green.

Learn More...

Prime Shapes in Natural Images

This paper provides evidence that about half of all the regions in segmented images can be classified as one a few simple shapes. Using three segmentation algorithms, three different image databases, and two shape descriptors, we empirically show that shapes such as triangles, squares, and circles are observed, up to an affine transform and at a much higher rate than random shapes. This result has potential value in applications such as scene understanding, visual object classification, and matching because qualitative shapes can be used as features. We show an application in scene categorisation based on what might be called 'bag of shapes' .

Fig. Segmented regions classified by prime shapes obtained from MIT database. Example: the segmented torso of a man is classified as an ellipse/circle.

Learn More...

Non-Photorealistic Rendering in Chinese Painting

In this thesis, we present an automatic framework to convert an input image to a stylized painting with Chinese wash painting effect automatically. Some traditional algorithms are improved and some new algorithms are developed as well.

LEFT: Original input image. MIDDLE: Our processing result. RIGHT: Real handmade Chinese painting

Learn More...