Weak Hypotheses and Boosting for Generic Object Detection and Recognition

Abstract

In this paper we describe the first stage of a new learning system for object detection and recognition. For our system we propose Boosting [5] as the underlying learning technique. This allows the use of very diverse sets of visual features in the learning process within a common framework: Boosting -together with a weak hypotheses findermay choose very inhomogeneous features as most relevant for combination into a final hypothesis. As another advantage the weak hypotheses finder may search the weak hypotheses space without explicit calculation of all available hypotheses, reducing computation time. This contrasts the related work of Agarwal and Roth [1] where Winnow was used as learning algorithm and all weak hypotheses were calculated explicitly. In our first empirical evaluation we use four types of local descriptors: two basic ones consisting of a set of grayvalues and intensity moments and two high level descriptors: moment invariants [8] and SIFTs [12]. The descriptors are calculated from local patches detected by an interest point operator. The weak hypotheses finder selects one of the local patches and one type of local descriptor and efficiently searches for the most discriminative similarity threshold. This differs from other work on Boosting for object recognition where simple rectangular hypotheses [22] or complex classifiers [20] have been used. In relatively simple images, where the objects are prominent, our approach yields results comparable to the state-of-the-art [3]. But we also obtain very good results on more complex images, where the objects are located in arbitrary positions, poses, and scales in the images. These results indicate that our flexible approach, which also allows the inclusion of features from segmented regions and even spatial relationships, leads us a significant step towards generic object recognition.

A Statistically Selected Part-Based Probabilistic Model for Object Recognition

“…We compared our result to the state of the art results from [6] and [12]. Table 1 summarizes the recognition accuracy at the equal ROC points (point at which the true positive rate equals one minus the false positive rate) of our different approach: no part selection with PCA, part selection with PCA, part selection with 2D PCA and results from other recent methods.…”

A life-long learning vector quantization approach for interactive learning of multiple categories

“…In contrast to this, discriminant models directly learn the mapping from x to t c based on a decision function Φ(x) or estimate the posterior class probability P (t c |x) in a single step (Ng & Jordan, 2001). Common approaches for this group of categorization models are based on support vector machines (Heisele et al, 2001), boosting (Viola & Jones, 2001;Opelt et al, 2004) or SNOW (Agarwal et al, 2004). Such discriminant models tend to achieve a better categorization performance compared to generative models if a large ensemble of training examples is available (Ng & Jordan, 2001).…”

A Boundary-Fragment-Model for Object Detection

“…Note that the amount of supervision varies over the methods where e.g. [26] use labels and bounding boxes (as we do); [2,3,12,22] use just the object labels; and Sivic et al [25] use no supervision. It should be pointed out, that we use just 50 training images and 50 validation images for each category, which is less than the other approaches use.…”

“…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 [10], Bar-Hillel et al [3] and Opelt et al [22] simply use a "bag of visual words" model (with no geometrical relations between the parts at all), Agarwal & Roth [1], Amores et al [2], and Vidal-Naquet and Ullman [27] use quite loose pairwise relations, whilst Fergus et al [12] 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).…”