Maximum-likelihood approximate nearest neighbor method in real-time image recognition

Abstract

We explore the problems of classification of composite object (images, speech signals) with low number of models per class. We study the question of improving recognition performance for medium-sized database (thousands of classes). The key issue of fast approximate nearest-neighbor methods widely applied in this task is their heuristic nature. It is possible to strongly prove their efficiency by using the theory of algorithms only for simple similarity measures and artificially generated tasks. On the contrary, in this paper we propose an alternative, statistically optimal greedy algorithm. At each step of this algorithm joint density (likelihood) of distances to previously checked models is estimated for each class. The next model to check is selected from the class with the maximal likelihood. The latter is estimated based on the asymptotic properties of the Kullback-Leibler information discrimination and mathematical model of piecewise-regular object with distribution of each regular segment of exponential type. Experimental results in face recognition for FERET dataset prove that the proposed method is much more effective than not only brute force and the baseline (directed enumeration method) but also approximate nearest neighbor methods from FLANN and NonMetricSpaceLib libraries (randomized kd-tree, composite index, perm-sort).

Image retrieval of wool fabric. Part III: based on aggregated convolutional descriptors and approximate nearest neighbors search

“…In the case of massive high-dimensional datasets, the ENN search algorithm is costly to search the exact neighbors due to the phenomenon that is called the curse of dimensionality. 30 Nevertheless, different ANN search algorithms were proposed to efficiently search sufficiently nearby objects to a query object. Annoy is an open-source library of Spotify and an empirically engineered algorithm that searches the ANNs in high-dimensional space, which has been recognized as one of the best ANN libraries.…”

Research on image sensitive information recognition based on machine learning

“…After the color features of the image are extracted, shape features in the image are further developed. In this process, the edges of things in the image need to be detected first [16].…”

A Distributed Approximate Nearest Neighbor Method for Real-Time Face Recognition

“…In all machine learning problems, one of the most important and challenging steps is feature extraction. [3] In face recognition and generally image recognition problems, this process is more challenging due to the effect of lighting and complex background variability [5] and also variability of objects presented in images [4]. There are many ways for feature extraction such as "SIFT 1 " [6] which was applied in [7], "SURF 2 " [8] applied in [9], and "HOG 3 " [10] applied in [11].Based on the application, we can decide which one to use.…”

“…There are many ways for feature extraction such as "SIFT 1 " [6] which was applied in [7], "SURF 2 " [8] applied in [9], and "HOG 3 " [10] applied in [11].Based on the application, we can decide which one to use. After extracting the feature of all images, it is time to use machine learning algorithms [3] to classify the input image. Face recognition is both important and challenging tasks especially in working with big data-set [12].…”