Analysis of camera image processing principles-anti-noise, zoom, and so on

Http://blog.csdn.net/qikaibinglan/article/details/5882880

1.1 The increase of Ag in anti-noise processing will inevitably increase the noise. In addition, if the light is darker and the exposure time is too long, it will increase the number of noise points (from the perspective of a digital camera, this is mainly because of prolonged exposure, rising temperature of photosensitive elements, and increasing noise of photosensitive elements due to current noise. The defects of photosensitive elements are also one of the sources of noise and even bad points. Therefore, sensor integration or backend ISP usually have noise reduction settings. 1.1.1 The startup time is based on the noise, mainly because the noise reduction function needs to be enabled after Ag or exptime exceeds a certain value. Therefore, it is usually necessary to determine the threshold values of these two parameters. From the following method, we can see that the noise reduction trend is accompanied by a reduction in image quality, so the noise reduction function is enabled too early, noise reduction not only increases the burden on the processor or ISP, but may also be counterproductive. When the noise reduction function is enabled too late, it cannot be used as needed. 1.1.2
How can we determine whether a point is a noise? We start from how people identify noise. For human eyes, it is almost impossible to determine a noise, that is, the brightness or color of this point is significantly different from that of most points on the edge. From the mechanism of noise generation, the color exception should always be accompanied by the brightness exception, and the workload for handling brightness exceptions is smaller than that of color exceptions, therefore, when the difference between the brightness of a vertex and the brightness of the surrounding vertex is greater than a threshold value, the sensor ISP determines that the sensor is a noise. The processing method is usually to take the mean value for the surrounding points to replace the original value. This method does not increase the amount of information, similar to a fuzzy algorithm. For high-end digital cameras, there is a strong image processing chip, and it is also possible to determine whether there are more complex algorithms for processing and determination. For example, brightness and color are used as the standard to determine the noise, and a greater amount of interpolation algorithm is used for compensation. For the inherent bad points and noise of sensor, block the data (Nikon does this, so should other vendors) and so on. 1.1.3
For mobile phone sensor, the effect of noise reduction is very large. I personally think it should be very limited. After all, compared with digital cameras, mobile phone sensor has a small lens and a small amount of light, therefore, the reference AG is bound to be greater than the camera gain (for example, equivalent to the iso800 level of a common digital camera), in order to obtain the same brightness, therefore, the impact of current noise is also much greater. In this way, even in the best case, there will be a lot of noise, and the data itself will fluctuate a lot, which will inevitably make us look at the phone photos, if the average operation reduces the noise, the image will also become blurred. Therefore, the threshold value for determining the noise on the mobile phone will be set to a relatively high level, so as not to blur the overall image. In this way, the data itself is poor, and the noise reduction standards are also reduced, resulting in poor overall results. 1.2
Digital zoom digital zoom can be in two forms: one is to use the interpolation algorithm to perform Interpolation on the image and expand the image size to the desired specifications, this is not ideal. Especially when using a mobile phone, the data produced by the camera on the mobile phone has a great deal of noise. If interpolation is performed, the image is almost unusable. In fact, even digital zoom functions of digital cameras do not have much practical value. If the interpolation algorithm is not supported by hardware, it must be implemented at the application layer. This method is used for digital zoom on a platform of our company. Second, it is actually a pseudo-digital zoom mode. when the camera is not in the maximum resolution format, for example, when a 1.3 million pixel sensor uses a 640*480 type photo, still set the sersor to work at a resolution of 1280*960, and then collect the central part of the image to get a 640*480 photo, this doubles the size of the captured object on the mobile phone. There are also many mobile phones that use this digital zoom mode. This method almost does not require additional algorithm support and has no impact on image quality. The disadvantage is that it can be used only when it is small. In addition, the digital zoom-and-zoom shooting function can be implemented in DV mode. (This should be a selling point. For DV, this digital zoom is still of practical significance)

To use this zoom mode, the driver must support the adaptive wing function to obtain the necessary sensor image data.
1.3 flash suppression
1.3.1 what is pop-up
The common optical sources such as incandescent lamps, fluorescent lamps, and Z lamps used in daily use work directly with 220/50Hz alternating current, and change each time 50 times in plus or minus half weeks per second, as a result, the light flashes 100 (50 × 2) times in one second, and the voltage on the market is unstable, in this way, the so-called "Intermittent" occurs ".
The following table shows the fluctuations in light intensity of several light sources:

 

Because human eyes have certain lag and adaptability to light intensity changes, the brightness changes of the light source are usually invisible. But it still increases eye fatigue. Therefore, there will be so-called non-frequency flash sales in the market.
1.3.2 suppression of frequent flashes
For camera sensor, there is no human eye lag or adaptation process, so it is sensitive to changes in the brightness of the light source. If no suppression is performed, bright changes may flash in preview and DV modes.
How can this problem be solved? Taking into account the periodicity of the flash, the accumulated Brightness Value of the light source within a period should be roughly the same. Therefore, if the time to control the exposure is an integral multiple of the flash cycle, then, the brightness of each frame is basically the same, which can effectively suppress the effect of frequent flashes on the brightness of the image.
Therefore, in the Automatic exposure mode, the sensor will adjust the exposure time to an integer multiple of its cycle based on the frequency of frequent flashes. Because the AC frequency varies across regions, there are 50Hz/60Hz.
When setting the relevant sensor registers, calculate the number of clock cycles corresponding to the frequency and resolution of the current and sensor.