The studies of image processing and computer vision have been attempting to simulate human visual perception for decades to extract features and recognize objects from the images. Within these domains, the current work aims at improving feature extraction by employing feedback process from predictions of model-based features or motions to optimize signal-based features. The motion predictions utilize symbol-based features to employ feedback process. We represent 3D circles as model-based features in the current work. The 3D Circle models are computed with a sophisticated approach via 2D ellipses. An ellipticity-check function ensures that invalid 3D circle hypotheses are eliminated. A feedback mechanism is proposed that spatially propagates 3D symbol-based features or 3D model-based features to signal-based features via a three dimensional Gaussian function. Moreover, the updated signal triggers the re-creation of features. Such processes form a loop, which is called “signal-symbol loop”. Both artificial and real sequences demonstrate the performance of the feedback mechanism.