Current consumer depth sensors produce depth maps that are often noisy and lack sufficient detail. Enhancing the quality of the 3D depth data obtained from compact depth Kinect-like sensors is an increasingly popular research area. Although depth data is known to carry a signal-dependent noise, the state-of-the-art denoising methods tend to employ denoising techniques which are independent of the depth signal itself. In this paper, we present a novel adaptive denoising filter to enhance object recognition from 3D depth data. We evaluate the performance of our proposed denoising filter against other state-of-the-art filters based on the enhancement of object recognition accuracy achieved after denoising the raw data with each filter. In order to perform object recognition from depth data, we make use of Differential Histogram of Normal Vectors (DHONV) features along with a linear SVM. Experiments show that our proposed filter outperformed the state-of-the-art de-noising methods.