Increasing the accuracy of feature evaluation benchmarks using differential evolution

Abstract

Abstract-The accuracy evaluation of image feature detectors is done using the repeatability criterion. Therefore, a wellknown data set consisting of image sequences and homography matrices is processed. This data serves as ground truth mapping information for the evaluation and is used in many computer vision papers.An accuracy validation of the benchmarks has not been done so far and is provided in this work. The accuracy is limited and evaluations of feature detectors may result in erroneous conclusions.Using a differential evolution approach for the optimization of a new, feature-independent cost function, the accuracy of the ground truth homographies is increased. The results are validated using comparisons between the repeatability rates before and after the proposed optimization. The new homographies provide better repeatability results for each detector. The repeatability rate is increased by up to 20%.

High-Resolution Feature Evaluation Benchmark

“…We make use of the RAW image data from [12]. In [12], the benchmark is created using subsampled images of size 1536 × 1024 (1.5 megapixels).…”

“…An improved homography benchmark is provided in [12] with image resolutions of 1.5 megapixels per image. In addition, the accuracy of the Mikolajczyk benchmark is slightly increased using a dense image representation instead of image features.…”

“…We use the RAW camera data from the images of the data set [12]. The proposed technique exploits the ground truth data from [12] for initializing an evolutionary optimization for the computation of ground truth homographies between image pairs with resolutions of up to 8 megapixels.…”

“…The proposed technique exploits the ground truth data from [12] for initializing an evolutionary optimization for the computation of ground truth homographies between image pairs with resolutions of up to 8 megapixels. This technique is called homography upscaling.…”

“…Section 3 shows the accuracy results of the benchmark compared to [12] and the feature evaluation using the repeatability criterion. In Section 4, the paper is concluded.…”