| Contour detection aims to obtain the true contour information of the target from complex scenes.It is an important basis for advanced computer vision tasks such as image segmentation and target recognition.From a biological point of view,the reason why human visual system is able to quickly obtain effective contour features from complex images is mainly that it is a multi-channel,multi-hierarchical,serial and parallel coexistence complex structure.By drawing on the information response processes of neuron cells of different properties in the visual system,it helps to improve the robustness of the bionic contour detection algorithm,thereby deepening the understanding of targets in natural images and providing the more reliable and realistic input characteristics for the subsequent more advanced visual tasks.Aiming at the problem that the traditional detection algorithm does not sufficiently extract the contour features of the target in complex scenes,based on the dynamic processing mechanism of Y-cells in the primary visual cortex to different visual information,this paper proposes a contour detection algorithm based on the dynamic response mechanism of Y-cells receptive field.Studies have shown that the response of central classical receptive field region of Y-cells in the primary visual cortex to visual information can be modulated by the inhibitory kernels that simulate the surrounding non-classical receptive field,and the dynamic response of the receptive field of neuronal cells plays an extremely important role in the process of capturing the contour features of the target by the visual system.Inspired by this physiological characteristic,this paper recalculates the weights of different positions of the Gabor filter in a limited size based on the local feature information of the target image,combines the optimal response direction of the classical receptive field,and the log function is used to modulate the Gaussian function to model the suppression characteristics of non-classical receptive field regions,which effectively strengthens the protection of weak contour features and the suppression of texture information.The experimental results show that the F-Score(ODS)of the proposed algorithm on the BSDS dataset and the NYUD dataset are 0.63(BSDS300),0.64(BSDS500),and 0.55(NYUD).Compared with the existing bionic contour detection algorithms,the proposed algorithm not only strengthens the protection of weak contour features,but also achieves better suppression of texture information.Aiming at the problem of insufficient feature extraction in the existing single-channel and few-levels detection algorithm inspired by the response characteristics of X cells,based on the hierarchic feedforward mechanism of the X and Y parallel information response channels in the visual pathway,this paper proposes a contour detection algorithm based on the spatial summation characteristics of the receptive field of X-and Y-cells.Research shows that different information channels,different levels of information response processes,and efficient information integration mechanisms in the visual pathway are important guarantees that the visual system can quickly capture the contour features of the target.Inspired by this physiological characteristic,this paper models the X-and Y-parallel response channels of the retina/lateral geniculate nucleus to the visual cortex,combines the luminance contrast of grayscale images to integrate different properties feature response,and enhances the texture suppression of the output feature image by introducing a sparse coding mechanism.The experimental results show that the F-Score(ODS)of the proposed algorithm on the BSDS dataset and the NYUD dataset are 0.63(BSDS300),0.65(BSDS500),and 0.56(NYUD).Compared with the existing bionic contour detection algorithms,the proposed algorithm improves the problem of the loss of some contour features in the information processing process,further strengthens the protection of contour features and the suppression of texture information. |
PDF Full Text Request