Vector Quantizing Feature Space with a Regular Lattice

Abstract

Most recent class-level object recognition systems work with visual words, i.e., vector quantized local descriptors. In this paper we examine the feasibility of a dataindependent approach to construct such a visual vocabulary, where the feature space is discretized using a regular lattice. Using hashing techniques, only non-empty bins are stored, and fine-grained grids become possible in spite of the high dimensionality of typical feature spaces. Based on this representation, we can explore the structure of the feature space, and obtain state-of-the-art pixelwise classification results. In the case of image classification, we introduce a class-specific feature selection step, which takes the spatial structure of SIFT-like descriptors into account. Results are reported on the Graz02 dataset.

Lost in quantization: Improving particular object retrieval in large scale image databases

“…Oxford Paris Fixed Quantization [18] 0.164 HKM [14] (1 level) 0.422 0.401 HKM [14] (2 level) 0.410 0.340 Hard [15] 0.614 0.403 Soft 0.673 0.494 Table 3. Comparison of soft-and hard-assigned vocabularies with the hierarchical k-means and fixed quantization.…”

“…The fixed quantization method [18] performs significantly worse as it is not suitable for specific object retrieval. Table 2.…”

“…no spatial ranking or query expansion) using the soft-and hard-assigned vocabularies with our implementation of hierarchical k-means (HKM) of [14] and fixed quantization of [18]. In the case of HKM we used the L 2 distance with 1 and 2 levels of hierarchical scoring (L 1 was found to degrade performance on this dataset [15]).…”

“…The experiments of [15] demonstrated that vector quantization obtained by this Approximate k-means (AKM) is superior to HKM. A fixed quantization method (complexity O(N )) was suggested in [18].…”

Fast Motion Consistency through Matrix Quantization

“…Vector quantization is commonly employed to reduce the computation time required to search for the nearest neighbors of a vector or to directly look up the result of a function on that vector [11,13,14,19]. The mapping of continuous feature spaces into a dictionary of prototypes results in more than just a computational speedup; it changes the way we view the problem and enables us to use fundamentally different approaches to solving the problem [6,20].…”