Target recognition and angle coarse estimation algorithm using space sparse coding

Abstract

The invention provides a target recognition and angle coarse estimation algorithm using space sparse coding. The algorithm includes the following steps that: the images of uniformly-spaced angles (15 degrees) of different targets are obtained so as to be adopted as a training set, the space segments of each image are screened according to standard deviation; whitening and PCA-combined pre-processing is performed on the segments in each independent image; the space segments are utilized to train the dictionary (sub-dictionary) of each target; after useless bases are removed from each sub-dictionary, the sub-dictionaries are merged integrally into a big dictionary, the big dictionary is utilized to re-obtain the sparse encoding coefficients of the fragments of the images in the training set, and the number of times of the use of bases in the big dictionary by the segments in each image is calculated, and is adopted as the feature vector of each training image; and finally, the correlation coefficient of the number of times of the use of the bases in the big dictionary by a test target image and the feature vectors of each image in the training set is calculated, so that target classification and angle coarse estimation are realized.

Description

technical field [0001] The invention relates to a digital image processing technology, in particular to a target recognition and angle rough estimation algorithm using space sparse coding. Background technique [0002] When the human eye recognizes and classifies objects, there are several parameters: color, shape, position, attitude, lighting conditions, observation points, interference or noise distribution, etc. In the context of big data, how to effectively abstract these parameters has become the primary problem of object recognition and classification. Sparse representation is currently a more effective method to deal with this problem. [0003] For computer vision applications, traditional methods include DCT, wavelet, etc. The above methods aim to use a large number of images to train an over-complete dictionary and then perform sparse coding on the target image. The obtained dictionary is pre-set, and it is very difficult to manually set a good dictionary. In addit...

AI Technical Summary

Problems solved by technology

However, PCA has strict requirements on the "base" in the dictionary, which must be strictly orthogonal, which limits the flexibility of problem solving. Sparse expression comes from this, and sparse coding expresses the original signal as a linear combination of dictionary elements.

Images