visionfans's research lab

关于codebook的不同用法，摘自[1]：

The methods differ on the details of the codebook, but more fundamentally they differ in how strictly the geometry of the configuration of parts constituting an object class is constrained.

For example, Csurka et al. [2], Bar-Hillel et al. [3] and Opelt et al. [4] simply use a "bag of visual words" model (with no geometrical relations between the parts at all), Agarwal & Roth [5], Amores et al. [6], and Vidal-Naquet and Ullman [7] use quite loose pairwise relations, whilst Fergus et al. [8] have a strongly parametrized geometric model consisting of a joint Gaussian over the centroid position of all the parts. The  approaches using no geometric relations are able to categorize images (as containing the object class), but generally do not provide location information (no detection). Whereas the methods with even loose geometry are able to detect the object's location.

从图像中提取出visual fragments之后，如何利用呢？

1. 无组织的利用：bag of visual words [2,3,4]
2. 考虑visual fragments之间的组织关系，怎么表示它们之间的关系呢，待详读[5,6,7,8]。

Referecnes:

[1] Opelt, A.; Pinz, A. & Zisserman, A. A Boundary-Fragment-Model for Object Detection European Conference on Computer Vision, 2006, 575-588

[2] G. Csurka, C. Bray, C. Dance, and L. Fan. Visual categorization with bags of keypoints. In ECCV04. Workshop on Stat. Learning in Computer Vision, pages 59-74, 2004.

[3] A. Bar-Hillel, T. Hertz, and D. Weinshall. Object class recognition by boosting a part-based model. In Proc. CVPR, volume 2, pages 702-709, June 2005.

[4] A. Opelt, M. Fussenegger, A. Pinz, and P. Auer. Weak hypotheses and boosting for generic object detection and recognition. In Proc. ECCV, pages 71-84, 2004.

[5] S. Agarwal, A. Awan, and D. Roth. Learning to detect objects in images via a sparse, part-based representation. IEEE PAMI, 26(11):1475-1490, Nov. 2004.

[6] J. Amores, N. Sebe, and P. Radeva. Fast spatial pattern discovery integrating boosting with constellations of contextual descriptors. In Proc. CVPR, volume 2, pages 769-774, CA, USA, June 2005.

[7] M. Vidal-Naquet and S. Ullman. Object recognition with informative features and linear classi¯cation. In Proc. ICCV, volume 1, pages 281-288, 2003.

[8] R. Fergus, P. Perona, and A. Zisserman. Object class recognition by unsupervised scale-invariant learning. In Proc. CVPR, pages 264-271, 2003.

Some notes related to contour grouping and matching

The problem of computational perceptual grouping received considerable attention before the advent of appearance-based recognition, when object models were typically shape-based and image features were typically contour-based.[1]

Moreover, while object databases were rather small, it was generally assumed that a linear search of a database, i.e., matching the image features against each model in succession and choosing the best-matching model, was an unacceptable strategy, for it did not scale to very large databases. In an effort to achieve sublinear scaling, much effort was devoted to the problem of object indexing, i.e., using a set of image features to query the database for candidate objects that might account for the image features.[1]

拿模型一个一个去匹配的方法已经过时了，因为现在的图像库都很大，计算上不现实了。现在主要是研究如何更好的进行object indexing。

grouping was based not on object-level prior knowledge, but rather on mid-level (object-independent) prior knowledge. Such grouping was essential, since local contour features were highly ambiguous, and without grouping them into more discriminative structures, effective indexing into large databases was problematic.[1]

Grouping所组织出来的是mid level的信息，是有用的。

References:

[1] Sala, P. & Dickinson, S. J. Contour Grouping and Abstraction Using Simple Part Models European Conference on Computer Vision, 2010, 603-616