Tilt correction of images used for surveillance
Pallav Nandi Chaudhuri, Samriddha Dey, Subhadip Bhattacharya · International Journal of Modern Trends in Engineering and Research · 2016
Tilt correction of images is an important step of different object recognition algorithms. In this paper we proposed a method to correct the tilt in images using linear symmetry feature. Linear symmetry feature is extensively used in OCR technology and finger print recognition. We propose to use this in a generalized way to recognize the tilt in surveillance images. Long and sharp vertically or horizontally oriented edges in the image are used as reference for tilt correction. Keywords — Gaussian filter, linear symmetry, noise removal, orientation tensor, skew detection, skew estimation and removal, Tilt correction. I. INTRODUCTION In the images captured by surveillance cameras the tilt can occur due to misalignment of the camera as it is dynamic handled or due to the faulty alignment of objects where the objects deviated from their original inclination. The amount of tilt can be recognized using the proposed method. The orientation of the major objects in the image provides for the perception of tilt in the image. The orientation of the major objects can be calculated by computing the orientation tensor of the image which in turn can be predicted by calculating the linear symmetry feature present in the image (1). Linear symmetry feature is defined as the lines present in the image along which the direction of pixel value change is consistent. It is computed by calculating the second order complex moment (2).The orientation tensor thus estimated will give the perception of the amount of skew present in the image. The linear symmetry feature is extensively used in OCR technology (3) and finger print recognition algorithms (4). In this paper we tried to extend this idea to surveillance images which are more generalized in nature. The occurrence of tilt in surveillance images are more compared to OCR and finger print images, as they are captured in dynamic environments. Hence tilt correction becomes an important part at the time of processing of these images. We propose to use orientation tensor to calculate the angle of tilt in this case based on linear symmetry feature. As the surveillance images are usually captured in human occupied areas and jungles, the presence of houses, buildings, long trees etc. contribute long and sharp vertically or horizontally oriented edges to the image. These edges can be used as reference for tilt correction. Using our proposed method we have recognized the orientation angles of these long edges to calculate the amount of tilt in the image.