Visual servoing is a control strategy that uses visual feedback from cameras to control the motion of a robot or a system. Image-based visual servoing relies on image processing and computer vision algorithms to detect and track image features, incorporating them directly in the control loop. That approach considers that there is a map, also known as interaction jacobian, between feature motion and camera pose based on the camera's intrinsic and extrinsic parameters. Although there are calibration techniques to compute the camera's parameters, they can become error-prone or need online changes, especially in unstructured scenarios. Some examples that could happen are when a task requires image zoom, the camera presents lens distortion, or its sensor has temperature sensitivity. Uncalibrated visual servoing studies aim to estimate the interaction jacobian using environment information and the measurement of features displacement, generally at run-time, with the help of an estimator, such as the Kalman filter. While most studies approximate the estimation uncertainty to a gaussian distribution, the environment in which the robot actuate could present more challenging characteristics. In that case, target occlusion, reflection, or similar appearance to other image objects can confuse the computer vision algorithm leading to outliers in the feature extraction. If not treated correctly, these errors may poor the performance of the visual servoing controller, or even affect its convergence. The maximum correntropy criterion can take advantage of the statistical properties of non-gaussian random variables. Thus, this work proposes a thesis theme study on dealing with non-gaussian feature tracking noise, preserving its statistical properties through the maximum correntropy criterion applied to the Kalman filter.