One of the challenges in ultrasound bone imaging is related to the specular reflection in the soft tissue-bone interface. Since the transmit ultrasound beam is not necessarily perpendicular to the bone surface, the reflected ultrasound signals from bones are often tilted in the channel data. Even though the amount of tilting may not be large, this can still result in cancellation of the bone signals during beamforming, weakening the bone interfaces. In this paper, we have developed a novel beamforming method which applies directional Log-Gabor filtering in channel data prior to the beamforming. The channel data are first delay compensated, and then the directional LogGabor filter are used to enhance signals within +/-15 degrees of horizontal direction (6 dB FWHM). The filtered channel data are then masked based on tensor based-image analysis to suppress noise. Vertebrae phantom studies (10 data sets) and in vivo volunteer studies (30 data sets) show the average contrast ratio (CR) for the bones is improved from 0.57 in delay and sum beamforming (DAS) to 0.91 in the proposed method. Weak bone interfaces such as the spinal process can also be visualized.