Rain streak removal using emboss and spatial-temporal depth filtering technique in video keyframes

Dynamic weather elements such as rain cause complex visual appearance, because rain consists of spatially distributed drops falling at high velocities. The continuous movement in spatio-temporal depth causes the distraction in the motion in a video sequence. Each drop when falling in high speed will create a streak motion blurred illusion based on the background intensity that reflects the environment creating higher intensity pattern in an image. In this thesis rain streak have been captured and isolated from the background scene by using an embossed filter algorithm designed to highlight the transparent rain streaks that cause the blur and distortion of the video, while the removing algorithm is based on a simple algorithm that correlates spatio-temporal and depth of an image into one technique. The removing method was based on the filtration process applied on the divided image blocks in the spatio-temporal depth technique controlled by an automatic steerable on noise availability in the image. The filter algorithm is based on an enhanced harmonic mean filter algorithm. Different techniques were applied conducted with different kernel structure. The image was clustered into layers,separating the RGB layers from the rain in one of the filters. In another filter a nested kernel was designed with the applicability of the main filter process. In another approach the image components are divided into two space criteria, conducting it in two different spaces and rejoined together. The kernel coefficients were automatically been set based on the noise availability in the image. The results obtained were satisfying, where the algorithm were able to detect and remove the rain streaks without losing the aliveness of the scene. The techniques presented here can be used in a wide range of applications including video surveillance, vision based navigation, video editing and video indexing.

Preview PDF FK 2012 75R.pdf Download (953kB) | Preview

Actions (login required)

View Item