EFFICIENCY EVALUATION OF A SPATIAL-FREQUENCY COVARIANCE DESCRIPTOR OF SMALL OBJECTS ON AERIAL IMAGES

Efficiency estimation of a spatial-frequency covariance descriptor of small objects in aerial images is provided. It is based on analysis of computational complexity, storage capacity, Riemann distance values, ROC and DET curves for real video sequences. A criterion of QR-algorithm of choosing of number of steps to compute generalized eigenvalues of covariance matrices is proposed. It is based on analysis of generalized eigenvalues histograms, probability of object correct detection in an aerial image and influence of scale factor and rotation angle of object on Riemann distance value. An estimation of discrimination ability of the spatial-frequency covariance descriptor based on analysis of AUC of ROC-curve is given. A criterion to choose the optimal threshold of object detection based on the Youden’s index and Equal Error Rate is defined. A range of threshold values to make decision of object detection based on the analysis of ROC- and DET-curves for real video sequences is determined. It provides a high probability of object correct detection at given probability of its false detection and optimal ratio between probabilities of type I errors and type II errors.

aerial image, spatial-frequency covariance descriptor, search of small objects, integral image, generalized eigenvalues, false positives and false negatives

1. Hu W., Li X., Luo W., Zhang X., Maybank S., Zhang Z. Single and multiple object tracking using Log-Euclidean Riemannian subspace and block-division appearance model. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2012, vol. 34, no. 12, pp. 2420–2440. Doi: 10.1109/TPAMI.2012.42

2. Yang C., Duraiswami R., Davis L. Efficient Mean-Shift tracking via a new similarity measure. IEEE International Conference on Computer Vision and Pattern Recognition, 2005, pp. 176–183. Doi: 10.1109/cvpr.2005.139

3. Baryskievic I., Tsvetkov V. Yu. Spatial-frequency covariance search for low-sized targets based on undecimated Haar lifting wavelet transform. Doklady Natsional’noi akademii nauk Belarusi = Doklady of the National Academy of Sciences of Belarus, 2014, vol. 58, no. 3, pp. 16–21 (in Russian).

4. Baryskievic I., Tsvetkov V. Yu. Stabilization of video sequence from board of a small UAV based on covariance search of reference points with prediction. Vestsi Natsyyanal’nai akademii navuk Belarusi. Seryya fizika-technichnych navuk = Proceedings of the National Academy of Sciences of Belarus. Physical-technical series, 2015, no. 1, pp. 106–111 (in Russian).

5. Ehsan S., Clark A. F., Rehman N .U., McDonald-Maier K. D. Integral images: Efficient algorithms for their computation and storage in resource-constrained embedded vision systems. Sensors, 2015, vol. 15, no. 7, pp. 16804–16830. Doi: 10.3390/ s150716804

6. Golub G. H., Van Loan C. F. Matrix Computations. London, Johns Hopkins University Press, 1996. 694 p.

7. Fawcett T. An introduction to ROC analysis. Pattern Recognition Letters, 2006, vol. 27, no. 8, pp. 861–874. Doi: 10.1016/j.patrec.2005.10.010

8. Schisterman E. F., Perkins N. J., Liu A., Bondell H. Optimal cut-point and its corresponding Youden index to discriminate individuals using pooled blood samples. Epidemiology, 2005, vol. 16, no. 1, pp. 73–81. Doi: 10.1097/01. ede.0000147512.81966.ba

9. Šimundić A.-M. Measures of diagnostic accuracy: basic definitions. Medical and Biological Sciences, 2008, vol. 22, pp. 61–65