Robust Analysis of Silhouettes by Morphological Size Distributions

Abstract

Abstract. We address the topic of real-time analysis and recognition of silhouettes. The method that we propose first produces object features obtained by a new type of morphological operators, which can be seen as an extension of existing granulometric filters, and then insert them into a tailored classification scheme. Intuitively, given a binary segmented image, our operator produces the set of all the largest rectangles that can be wedged inside any connected component of the image. The later are obtained by a standard background subtraction technique and morphological filtering. To classify connected components into one of the known object categories, the rectangles of a connected component are submitted to a machine learning algorithm called EXtremely RAndomized trees (Extra-trees). The machine learning algorithm is fed with a static database of silhouettes that contains both positive and negative instances. The whole process, including image processing and rectangle classification, is carried out in real-time. Finally we evaluate our approach on one of today's hot topic: the detection of human silhouettes. We discuss experimental results and show that our method is stable and computationally effective. Therefore, we assess that algorithms like ours introduce new ways for the detection of human in video sequences.

Boosting Shape Classifiers Accuracy by Considering the Inverse Shape

“…The first shape representation technique considered in this paper is the Cover By Rectangles (CBR) descriptor, introduced by Barnich et al [4], which decomposes an object in its largest rectangular components. According to the comprehensive taxonomy of da Fontoura et al [8], this is a region-based descriptor based on a region decomposition.…”

“…Barnich et al [4] encoded a rectangle by the coordinates of its centroid, its width, and its height. That encoding is inappropriate here as the centroid is undefined for rectangles expanding to infinity on two opposite sides.…”

“…In order to classify shapes based on the CBR representation, we adopt the object classification method of Barnich et al [4], for its simplicity. It has been proven to be efficient for the recognition of human silhouettes [4,21,22].…”

Object Descriptors Based on a List of Rectangles: Method and Algorithm

“…In [3] for example, the shape descriptor is based on the histograms of all the rectangle widths and heights. A subset of all maximal rectangles was successively used to discriminate human shapes from objects shapes in [2]. A granulometric curve is easily derived from γ k (X).…”

“…The advantages of having a list of rectangles are twofold: (1) it is simple to calculate granulometric curves once the list has been built because the list gathers all the information needed for the granulometry, and (2) it is possible to calculate, for each pixel, some statistics extracted from the list, like the size of the largest rectangles comprising that pixel. Descriptors based on a list of rectangles have proved successful for two classification tasks related to the analysis of human activities; they were used for gait recognition in [3], and for identifying human silhouettes in video scenes both in 2D [2] and in 3D [5].…”