Learning Active Basis Model for Object Detection and Recognition

By Ying Nian Wu, Zhangzhang Si, Haifeng Gong, Song-Chun Zhu. 2010

182 citations recorded by the source corpus.

This article proposes an active basis model, a shared sketch algorithm, and a computational architecture of sum-max maps for representing, learning, and recognizing deformable templates. In our generative model, a deformable template is in the form of an active basis, which consists of a small number of Gabor wavelet elements at selected locations and orientations. These elements are allowed to slightly perturb their locations and orientations before they are linearly combined to generate the observed image. The active basis model, in particular, the locations and the orientations of the basis elements, can be learned from training images by the shared sketch algorithm. The algorithm selects the elements of the active basis sequentially from a dictionary of Gabor wavelets. When an element is selected at each step, the element is shared by all the training images, and the element is perturbed to encode or sketch a nearby edge segment in each training image. The recognition of the deformable template from an image can be accomplished by a computational architecture that alternates the sum maps and the max maps. The computation of the max maps deforms the active basis to match the image data, and the computation of the sum maps scores the template matching by the log-likelihood of the deformed active basis.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Convolutional feature masking for joint object and stuff segmentation (2014, methodology): “To generate one compact combination from this candidate set, we adopt a procedure similar to the matching pursuit [23, 17].”
A Theory of Generative ConvNet (2016, background): “Sparsifying the connections will make the compositions more explicit and meaningful [37, 35].”
Learning FRAME Models Using CNN Filters (2015, methodology): “Such a scheme was used by (Hong et al. 2014) to learn codebooks of active basis models (Wu et al. 2010).”
Wasserstein Introspective Neural Networks (2017, background): “Before the deep learning era [17], generative modeling had been an area with a steady pace of development [1, 5, 54, 36, 49, 53].”
Mining And-Or Graphs for Graph Matching and Object Discovery (2015, background): “els [16, 34] focuses on geometric shapes and ignores textures.”
Sparse Representation Based Fisher Discrimination Dictionary Learning for Image Classification (2014, methodology): “Different from those class-specific DL methods (Ramirez et al. 2010; Yang et al. 2010c; Mairal et al. 2008b; Sprechmann and Sapiro 2010; Wang et al. 2012; Castrodad and Sapiro 2012; Wu et al. 2010), in FDDL the representation coefficients will also be made discriminative under the Fisher criterion.”
Learning Hybrid Image Templates (HIT) by Information Projection (2012, methodology): “In sparse coding and generative modeling, the active basis model [2] learns a sparse shape template composed of Gabor wavelet elements at different locations by information projection [21], [19].”
Unsupervised learning of compositional sparse code for natural image representation (2013, background): “[37] for modeling deformable templates formed by basis functions.”
Learning AND-OR Templates for Object Recognition and Detection (2013, methodology): “An ABT consists of a small number of Gabor basis elements (sketches) at select locations and orientations, where these basis elements are allowed to perturb their locations and orientations locally.”
Inducing wavelets into random fields via generative boosting (2016, unknown): “We compare our method with (a) the active basis model [23], (b) two-step EM [1], (c) k-means with HoG features [4].”
Semantic part segmentation using compositional model combining shape and appearance (2014, background): “There are lots of works in the literature on modeling object shape such as [25, 15, 28, 20].”
Learning Sparse FRAME Models for Natural Image Patterns (2014, methodology): “This leads to the deformable shared sparse coding first proposed in our previous work on active basis (Wu et al. 2010) Im = n∑ i=1 cm,i Bxi+Δxm,i ,si ,αi+Δαm,i + m, (17) where (Δxm,i ,Δαm,i ) are the perturbations of the location and orientation of the i-th basis function.”