Scale-invariant medial features based on gradient vector flow fields

Abstract

We propose a novel set of medial feature interest points based on gradient vector flow (GVF) fields [18]. We exploit the long ranging GVF fields for symmetry estimation by calculating the flux flow on it. We propose interest points that are located on maxima of that flux flow and offer a straight forward way to estimate salient local scales. The features owe their robustness in clutter to the nature of the GVF which accomplishes two goals simultaneously - smoothing of orientation information and its preservation at salient edge boundaries. A learning framework based on them, in contrast to classical edge-based feature detectors, would unlikely be distracted by background clutter and spurious edges, as these new mid-level features are shape-centered. We evaluate our scale-invariant feature coding scheme against standard SIFT keypoints by demonstrating generalization over scale in a patch-based pedestrian detection task

Shape centered interest points for feature grouping

“…SCIP features are tightly connected with the surrounding shape and thus offer the possibility to estimate a meaningful local scale as has been shown by us in [2]. The features are centered between at least two salient edges.…”

“…In this line of research wave propagation models based on the solution of the Eikonal equation in the image plane have been employed to derive features for shape matching tasks. Alternatively, GVF [1] fields have been recognized for the computation of medial features in cluttered scenes [2]. The GVF approach made many vision tasks amenable for cluttered environment.…”

“…This method can become unstable in the presence of noise and clutter. To overcome this we extend the approach of [2]. Using the normalized GVF field V N we calculate the scalar product between vector field V N and a disc with circularly inward pointing normals for different disc radii.…”

“…the Gradient Vector Flow (GVF) field [1]. Our algorithm is based on the medial feature extraction pipeline by Engel and Curio [2], which we have adopted here for the implementation of this novel feature grouping approach. After discussing related work in Section 2 we provide details of our implementation in Section 3.…”

Pedestrian detectability: Predicting human perception performance with machine vision

“…[14], [33]) with a Radial Basis Function (RBF) kernel as a regressor as it has proven to produce state-of-the-art results even on small and noisy datasets. We employed the freely available LIBSVM [11] implementation to train and test our Difference in mean color, standard deviation and the earth mover distances between difference types of histograms between the bounding box of the pedestrian and its context (see Figure 5 After resizing the image in the bounding box around the pedestrian to 100 × 50, we computed the flux flow F as described in [15]. mfThres is the number of pixels whose flux flow is above a threshold of 1 and represents the level of symmetry in this area.…”