Distortion Invariant Correlation Filters for Enhanced Target Recognition

Abstract

The target object recognition system faces important concerns including the sensitivity to the variations in the reference image. A fully invariant system helps in resolving difficulties in object detection when camera or object orientation and position are unknown. Correlation filters are the well-established means for target recognition tasks. Advanced correlation filter is an effective tool for target detection within a particular class. A combinational framework of correlation filters and logarithmic transformation had been previously reported to resolve this issue alongside catering for scale and rotation changes of the object in the presence of distortion and noise. The impact of different log bases on resulting correlation plane has been focused in our research. The contraction and expansion in the correlation peak have been detected for different situations which can be used for attaining enhanced tolerance to different distortions. The logarithmically transformed correlation filters have been tested for different circumstances to find an optimal log base for specific variance. The log base 10 can be used effectively to yield sharp peak in the presence of distortion. The log base 2 can be employed in logarithmic mapping to obtain better results for noisy target. Minimum Average Correlation Energy (MACE) filter yields sharp correlation peaks while considering the controlled correlation peak value. Difference of Gaussian (DOG) wavelet has been added at the preprocessing stage in proposed filter design that facilitates target detection in orientation variant cluttered environment. Logarithmic transformation is combined with a DOG composite minimum average correlation energy filter (WMACE), capable of producing sharp correlation peaks despite any kind of geometric distortion, occlusion and noisy target. The proposed correlation filter based mechanism has shown improved performance over some of the other variant correlation filters.