Research On Light Robustness Based Object Recognition

Object recognition is one classic technique in the field of computer vision,it aims at extracting the object region fully from target image.As one essential supporting algo-rithm of it,image enhancement algorithm usually extracts the image information from frequency domain and space domain.However,traditional algorithms usually process images using global processing strategy which couldn't process blocked object and im-age with significant light changing effectively.Aiming at these problems,human visual system based Retinex theory has obvious advantages in detail enhancement,color fidel-ity and dynamic range compression.On the foundation of it,our paper did some further research on Retinex algorithm and proposed robust image enhancement algorithm which performs well even for significant light changing image.The main research con-tent and innovative points are in the following:(1)Aiming at halo artifact problem based on center surround algorithm of Retinex the-ory,on the foundation of deep research for guide filter,our paper achieved guide filter based image enhancement algorithm which reduces the phenomenon of halo artifact around high-contrast image edge.(2)Aiming at the problem of balancing between detail enhancement and color fidelity,noise amplification after image processing,our paper improved the color recovery function and achieved guide filter based multi-scale Retinex image enhancement algorithm,which effectively balances the problem between image detail enhance-ment and color fidelity.(3)Aiming at the problem under the condition of light weak stability in object recogni-tion,our paper extracted the feature using SURF which performs robustly and ef-fectively.It abolished the traditional regular grid based extracting methods.The SURF based model reduce the useless image information(background),enhance the effective image information.At the same time,our paper combined above feature descriptor and spatial pyramid model(SPM)?SVM classifier,and optimized the histogram weight to achieve object recognition algorithm based on light robustness.(4)Design and construct a real-time object recognition framework for Android plat-form and light robustness,develop an object recognition software based on Android mobile device,apply the software to object recognition in real scene.At last,we did adequate experiments on database based on massive images.The experimental results indicate that our algorithm could solve the problem of noise pollu-tion and halo artifact around high-contrast image edge.Moreover,our algorithm could improve the color fidelity and recognition accuracy under the condition of significant light changing.Moreover,Android platform based object recognition software performs well in light robustness,timeliness and recognition accuracy.