Today consumers make increasingly high demands of the image quality. The first step in this direction was switching to high resolution (for example, 720x576 pixels). Although some time ago CIF resolution (360x288) was the common standard in video surveillance, full-size image is required in most cases nowadays. Yet switching to high resolution does not completely solve the problem of getting a high-quality image. Other factors besides resolution such as optical components, illuminators, video-capturing boards and compression algorithms have an effect on picture quality as well.

Contrary to analogue CCTV systems, in digital solutions it is possible to improve picture quality by using digital image processing. This allows obtaining high-quality images even when using not very good cameras or operating in unfavourable lighting conditions.

Vocord has developed a specialised hardware-software system that automatically adjusts image parameters in VOCORD Phobos and VOCORD Tahion systems.

Brightness and Contrast AGC (AGCBC)

One of the problems which always developers and users of digital video-monitoring systems face is the limited dynamic range of the recorded image. Loss of image details in some dark or bright areas is a characteristic feature for scenes with a wide dynamic range. Built-in standard options for automatic signal level control (AGC) do not solve the problem in full measure either.

Moreover, using standard AGC in cameras can lead to new problems because it maintains the average signal value, which causes image detorioration in some conditions. For example, a bright object appearing into the field of view leads to bad representation of dark parts of an image when using cameras with standard. This problem can be solved by manually setting the parameters but it diverts operator’s attention and as a result affects working with multi-camera security systems. This also makes the system settings along wth the final readings more subjective.

Fig.1. A night scene shot with “ideal” settings (left) and with “overfilling” (right)

AGCBC algorithm is designed for automatic maintenance of image brightness and contrast levels set by the operator by adjusting brightness and contrast settings of the analog video with the change of illumination conditions and composition of the observed scene.

The algorithm continuously calculates statistical values of image brightness and contrast to achieve this. With the decrease of calculated values the algorithm should increase them to specified meanings by changing video settings and vice versa.

Brightness Adjustment

AGCBC algorithm alone is not enough when setting regulation limits and observing highly-contrasting scenes with wide dynamic range where data loss is undesirable. A special software algorithm for brightness adjustment can solve this problem.

This algorithm is based on Retinex (formed from two medical terms: retina and cortex) - the psychophysiological model of human vision.

It is known that human identification of an object hardly depends on the level and spectral structure of object illumination. For example, the green apple colour is confidently identified practically in any lighting conditions. Moreover, human vision allows discerning the smallest details of a scene in a wide dynamic illumination range.

The process of image identification is a result not only of visual performance but also of the brain cortex activity which additionally processes visual information and corrects poor and intense lighting of different parts of the observed scene. The method of brightness adjustment developed on the basis of this model is built upon evaluation and the following removal of local illumination heterogeneity of objects of the observed scene. Using this method makes poorly lighted objects become more observable, object colour and details are restored well.

This method is intended most of all for high-contrast scenes with intense illumination and very shaded areas where local parts are usually indistinguishable. It is especially effective when working with excessively contrast archival videorecordings because it allows reconstructing indiscernible details.

Fig. 2. Brightness adjustment: the original picture (left) and a processed one (right)

A similar method is developed and used by NASA for processing of still images. It is worth noting that applying this method for low-contrast scenes is usually ineffective and can lead to some artifacts in the form of grey fog.

An optional mode for brightness adjustment is developed for high-contrast scenes - representation in the logarithmic scale.

Noise Reduction

The picture received from video monitoring systems always contains a noise component (Fig. 3, left). Its primary sources are:

the noise of a video camera, intensifying with the decrease of illumination of the observed object
the discretization noise, emerging during video digitization
compression algorithm noise

Moreover, it is necessary to distinguish the noise of brightness and color components of an image since their nature, influence on image perception and ways of removal are different.

The nonlinear filtering method

The method of nonlinear filtering is used to reduce noise of the brightness component of an image, it recognizes the object’s edges and retains them.

This method is based on segmenting an image into local zones with the similar level of brightness. Adjoining segment pixels for every pixel of an image are defined within the frame and, if possible, in time. Considerable image smoothing inside the selected segment is possible, while foreign objects are outside this area and as a result don’t cause blurring (Fig. 3, right).

Fig. 3. Noise reduction application: original image (left) and after applying the adaptive method for brightness component smoothing (right)

There are four different modes of noise reduction for the algorithm adaptation to the level and nature of noise:

Adaptive (Fig. 3, right) is the quickest and most effective mode for relatively static images because the spatio-temporal analysis of a sequence of images is used in this mode. When reproducing dynamically changing scenes the result can be insufficient in areas with intense motion.
Weak, medium and strong. In this mode only intraframe data is used. This method is universal but it is slower than the adaptive mode and the processing speed decreases with the amplification of noise reduction.

Color noise. Distortion of the color component of an image can appear in high-frequency noise and color ghosts near sharp edges of bright color objects (Fig. 4, left). Even the usage of noise reduction in the brightness component (Fig. 3, right) does not remove these artifacts especially noticeable in action.

A mode of color noise reduction was developed for the effective removal of these defects (Fig. 4, middle). Because of efficiency reasons the Hann 3x3 filter smoothing of the last two levels of the both color-difference components with reverse wavelet transformation is applied. It is moderately resource-intensive and quite suffecient in most cases.

The result of joint action of the brightness and color component noise reduction is shown on Fig. 4, right.

Fig. 4. Color noise reduction: original image (left), color component (middle), brightness and color components (right)

Sharpness Increase

Often a picture received from security video-monitoring system can be a bit fuzzy. The reason of this can be either bad shooting conditions or preceding stages of video processing (compression, noise damping, etc.) Sometimes in order to improve operator’s picture perception it is helpful to use the technology of picture sharpening.

The “unsharp mask” technology is used in this case. The main point of this method consists in increasing the picture contrast in the border lines of areas with different brightness (Fig.5) and a person subjectively considers the image to be more sharp.

Fig. 5. Principle of the "unsharp mask" algorithm

The effect of using the technology of picture sharpening in a VOCORD System is shown on Fig.6.

Fig.6. Picture sharpening: noise reduction (left) and noise reduction + image sharpening (right)

Deinterlacing

Analog television works in the interlace mode that is its every frame consists of two fields which contain even and odd lines. The frames shot by an analog camera are displayed without distortion on the analog monitor, which works in the interlace mode as well. Unfortunately, situation in the digital television is different: monitors with progressive scanning are used along with standard analog cameras and as a result the comb effect appears on borders of moving objects. The deinterlacing technology is applied to remove this effect from the image.

There are many different video deinterlacing methods differing in efficiency, application, processing speed, complexity, etc. In the VOCORD System the user can choose from three various deinterlacing methods:

Smoothing. This method is based on averaging brightness values of vertically adjoining pixels. It is applied concurrently to both even and odd lines, but it is not allowed to receive frames at double frequency. Its advantages are low system requirements and high processing speed independent of complexity and speed of objects’ movement within the frame. The disadvantages are: smearing comb into a semi-transparent loop after fast moving objects and image quality worsening with the loss of small details.
Interpolation. It is based on the interpolation of even lines’ values in order to receive the values of the adjoining odd lines and vice versa. This allows displaying fields of a videostream at double frequency. The advantages of this method are its high processing speed, absence of loops after moving objects and the retaining of small details. Its disadvantage is the possibility of image jitter during some types of motion.
Adaptive method. In this method every pixel is analysed in order to detect object movement nearby and afterwards it is decided whether it is necessary to interpolate the values of adjoining pixels to receive a new value of the current pixel or not. The adaptive method can be applied to even and odd lines independently so doubling of the output stream frequency is acceptable. Its advantages are: high frames quality, retaining of image details, low jitter and the absence of artifacts. The method’s only disadvantage is its high system requirements.

On Fig. 7 there are examples of using different deinterlacing methods for a typical interlaced image with fast and complex motion.

: a) Interlace

: b) Smoothing

: c) Interpolation

: d) Adaptive method