FDNR (Defog)

CCTV cameras are often installed outside, expected to operate 24/7 and not be affected by sunlight, rain, snow and fog and to reduce the adverse effect of the environment on the recorded image. Weather conditions significantly affect the quality of image recorded by the CCTV camera. Foggy weather conditions affect the image quality by reducing the contrast ratio, which in turn reduces the clarity and sharpness of the scene details recorded by the camera. The manufacturers often include many functions to enhance the quality of the recorded image.

In particular, the Defog function, also referred to as F-DNR is one of the key algorithms implemented among many functions developed for outdoor surveillance.

Fig. 1. Comparison of images with Defog function disabled (DEFOG OFF) and enabled (DEFOG ON)

The developments in digital technology introduced by the image processing algorithms made the Defog function more and more commonly used, even in the budget cameras. The technology adapts the distribution of information captured by the camera’s image sensor and improves image contrast and details to reduce information loss during post-processing.

There are two known digital defog algorithms, the non-model image enhancement and the model image recovery method. The non-model image enhancement method increases the contrast ratio to ensure a satisfactory subjective visual judgement by the user. The model image recovery method analyses the reasons why the image quality deteriorates and reverses the process to recover the original image quality.

The typical methods of non-model image enhancement include the histogram equalization, the filter transformation and the fuzzy logic based theory. The histogram equalization can be divided into the global histogram equalization and the partial histogram equalization. The global histogram equalization has a low computational cost but the enhancement of the image information is not sufficient. The partial histogram equalization enhancement works better but may cause block effects and increase noise. The filter transformation algorithm provides good image quality, but the computational cost and resource use too high for live surveillance. The defog effects based on fuzzy logic are usually insufficient.

All in all, the non-model image enhancement methods can improve the image quality to some extent but will not enhance the image quality in an effective way.

The model image recovery method includes the filtering method, the maximum entropy method and the degraded image function estimation method. The filtering method, such as the Kalman filtering, generally requires heavy computations. The maximum entropy provides high resolution but it is a highly complex non-linear algorithm with a complicated calculation process. The degraded image function estimation method is normally designed based on a certain physical model. This algorithm requires the images to be captured at different times as a reference to determine the parameters of the physical model and since this is a non-real-time method, it is difficult to use it in surveillance.

The largest CCTV manufacturers have developed a Defog function for live video. The technology is based on the atmospheric optics theory. The Defog function distinguishes the depth of field and fog density of different areas and uses filtering to get a clear and natural image.

Below are the examples of the Defog function.

Fig. 2. Before and after contrast ratio enhancement.

Fig. 3. Comparison of image details.

Fig. 4. Comparison of text in the scene.