Publications by Costantino Grana

In this paper we propose a novel image descriptor built by computing the covariance of pixel level features on densely sampled patches and encoding them using their covariance. Appropriate projections to the Euclidean space and feature normalizations are employed in order to provide a strong descriptor usable with linear classifiers. In order to remove border effects, we further enhance the Spatial Pyramid representation with bilinear interpolation. Experimental results conducted on two common datasets for object and texture classification show that the performance of our method is comparable with state of the art techniques, but removing any dataset specific dependency in the feature encoding step.

In this paper we address the task of learning how to segment a particular class of objects, by means of a training set of images and their segmentations. In particular we propose a method to overcome the extremely high training time of a previously proposed solution to this problem, Kernelized Structural Support Vector Machines. We employ a one-class SVM working with joint kernels to robustly learn significant support vectors (representative image-mask pairs) and accordingly weight them to build a suitable energy function for the graph cut framework. We report results obtained on two public datasets and a comparison of training times on different training set sizes.

In this paper, we propose a general approach for automatic segmentation, color-based retrieval and classification of garments in fashion store databases, exploiting shape and color information. The garment segmentation is automatically initialized by learning geometric constraints and shape cues, then it is performed by modeling both skin and accessory colors with Gaussian Mixture Models. For color similarity retrieval and classification, to adapt the color description to the users’ perception and the company marketing directives, a color histogram with an optimized binning strategy, learned on the given color classes, is introduced and combined with HOG features for garment classification. Experiments validating the proposed strategy, and a free-to-use dataset publicly available for scientific purposes, are finally detailed.

In this paper we propose to extend the well known graph cut segmentation framework by learning superpixel relations and use them to weight superpixel-to-superpixel edges in a superpixel graph. Adjacent superpixel-pairs are analyzed to build an object boundary model, able to discriminate between superpixel-pairs belonging to the same object or placed on the edge between the foreground object and the background. Several superpixel-pair features are investigated and exploited to build a non-linear SVM to learn object boundary appearance. The adoption of this modified graph cut enhances the performance of a previously proposed segmentation method on two publicly available datasets, reaching state-of-the-art results.

This paper reports a novel approach to deal with the problem of Object and Scene recognition extending the traditional Bag of Words approach in two ways. Firstly, a dataset independent method of summarizing local features, based on multivariate Gaussian descriptors, is employed. Secondly, a recently proposed classification technique, particularly suited for high dimensional feature spaces without any dimensionality reduction step, allows to effectively exploit these features. Experiments are performed on two publicly available datasets and demonstrate the effectiveness of our approach when compared to state-of-the-art methods.

In this paper we propose an approach for Document Layout Analysis based on local correlation features. We identify and extract illustrations in digitized documents by learning the discriminative patterns of textual and pictorial regions. The proposal has been demonstrated to be effective on historical datasets and to outperform the state-of-the-art in presence of challenging documents with a large variety of pictorial elements.

Techniques based on Bag Of Words approach represent images by quantizing local descriptors and summarizing their distribution in a histogram. Dierently, in this paper we describe an image as multivariate Gaussian distribution, estimated over the extracted local descriptors. The estimated distribution is mapped to a high-dimensional descriptor, by concatenating the mean vector and the projection of the covariance matrix on the Euclidean space tangent to the Riemannian manifold. To deal with large scale datasets and high dimensional feature spaces the Stochastic Gradient Descent solver is adopted. The experimental results on Caltech-101 and ImageCLEF2011 show that the method obtains competitive performance with state-of-the art approaches.

In this paper we propose to integrate the recently introduces ORB descriptors in the currently favored approach for image classification, that is the Bag of Words model. In particular the problem to be solved is to provide a clustering method able to deal with the binary string nature of the ORB descriptors. We suggest to use a k-means like approach, called k-majority, substituting Euclidean distance with Hamming distance and majority selected vector as the new cluster center. Results combining this new approach with other features are provided over the ImageCLEF 2011 dataset.

In this paper, the problem of automatic people removal from digital photographs is addressed. Removing unintended people from a scene can be very useful to focus further steps of image analysis only on the object of interest, A supervised segmentation algorithm is presented and tested in several scenarios.

In this paper we propose a large-scale Image annotation system for the Scalable Concept Image Annotation task. For each concept to be detected a separated classifier is built using the provided textual annotation. Images are represented as a Multivariate Gaussian distribution of a set of local features extracted over a dense regular grid. Textual analysis, on the web pages containing training images, is performed to retrieve a relevant set of samples for learning each concept classifier. An online SVMs solver based on Stochastic Gradient Descent is used to manage the large amount of training data. Experimental results show that the combination of different kind of local features encoded with our strategy achieves very competitive performance both in terms of mAP and mean F-measure.