• Chromatix


    1、Lens Rolloff Correction  透镜衰减矫正

    The Lens Rolloff correction takes into account the fact that,with some sensors,the center of the image is brighter than the edges.The Lens Rolloff correction brings up the edges of the image so that the image is uniform.The Lens Rolloff should be done in Chromatix first so that the correction can be applied to the raw images before going through optimiaztion.

    2、Color Correction Optimization 色彩校正优化

    2.1、Noise Weight

    The higher the noise weight,the more color accuracy will be sacrificed to reduce noise.

    2.2、Advaned Color Setting

    Saturation adjustment 饱和度调整 

    饱和度是指色彩的鲜艳程度,也称色彩的纯度。饱和度取决于该色中含色成分和消色成分(灰色)的比例。含色成分越大,饱和度越大;消色成分越大,饱和度越小。“色彩饱和度”越高,图片越鲜艳,反之越发灰。

    Reference color adjustment 标准色调整

    3、Gamma Optimization 伽马优化

    Gamma可能源于CRT(显示器/电视机)的响应曲线,即其亮度与输入电压的非线性关系。通常的感觉是,系统gamma高,图像会发暗,而校正后,画面变亮。

    * 当系统Gamma高(Gamma校正=1.0)时,看到的图像暗部影调丰富;反之,则亮部丰富。

    * Gamma校正设为2.5时看到的File Gamma=2.5的图片,和校正设为1.0时看到的File Gamma=1.0的图片相当。

    Gamma的变化带来亮度的变化。而单独改变某个通道的Gamma,则会则会带来色调(Hue)的变化。

     Gamma Option

    sRGB

    Ideal gamma function,inverse of gamma in an sRGB display,Piecewise,but close to power gamma with power λ=1/2.2

    Power Gamma

    Standard contrast,Compensate the gamma of sRGB display by taking into account image noise,color,contrast,and sensor nonlinearity simultaneously

    Custom gamma adjustment

    More flexible,Fine-tuning with reference gamma

    4、Color Conversation Optimisztion

    Noise weight and advanced color setting for each illuminant发光体

    5、ISO Gain 感光度增益

    ISO的数值每增加1倍,其感光的速度也相应的提高1倍。ISO越高,在黑暗的时候就能拍摄的更加明亮。感光度越低,仅对强光有反映,但成像细腻;感光度高,可在弱光下拍摄,但成像粗糙。ISO并不是越高越高,越高的ISO会带来非常多的噪点。

    Chromatix can help to generate an ISO gain value for ISO100 speed.You will need to input the f-number of the lens,the lux of the white patch,the exposure time,and the gain used to take the picture.Chromatix will calculate the gain for ISO100.

    6、Luma Target亮度指标

    The Luma Target Dialog will allow the user to input the desired luma target value after gamma correction.Chromatix will calculate the desired luma target before gamma correction so that the AMSS code can achieve the desired luma target.

    Setup Exposure Table

    • More gain may result in more noise,and the longer the sensor detects data(Line count/exposure time),the more likely it is thatblur will be encountered.
    • Linear AFR adjusts the frame rate to increase or decrease sensor exposure time when camera gain exceeds or drops below configured gain levels.
    • Lower frame rate allows longer exposure time,but increase skew and blurriness
    • Digital Assisted Gain option is to generate the exposure table automatically for the sensor that cannot support 3% sensitivity increments
    • Qualcomm usually provides the appropriate values for these items,Consult with Bayer Apps via the case system.

     Mesh Lens-Rolloff

    • More degree of correction gives better lens shading quality, but worse noise quality
    • You usually limit the maximum G and B gain 1.9,preferably around 1.7
    • Since VF and snapshot use different rolloff tables ,there is a color shift between them. The more a rolloff is made, the more the color shift will be.
    • The first thing to do after collecting Macbeth data is to check if the image was overexposed.If the lines are not fairly straight,the images should be retaken.

    •  In case there is no Low-Light Lens-Rolloff Raw image, users can use TL84 lens rolloff RAW image with 50% correction.

       It is recommended to use the same Lens-Rolloff image and the same Degree of Correction with Calculate Mesh Lens-Rolloff for Preview and TL84

     Color Correction

    •  Qualcomm recommends that you use deltaE*94 Color Difference Mode
    • The bigger noise weight gives better noise quality,but worse color delta
    •  It is recommended that the diagonal elements from left to right of Color Correction Matrix be between 1.0 and 1.9
    •  deltaE*ab Generally provides better color saturation,while detaE*94 provides better balance between the R,G,and B channels
    •  Place Macbeth markers on processed image and select Test->Color Difference Test(good way to decide whether to use E*ab or E*94)
    • Qualcomm recommends that you use deltaE*94 Color Difference Mode
    •  In general,the larger the diagonal coefficients,the larger crosstalk among RGB channels,the more noise is amplified

      It is recommended that the amcmapcp elements of Output be between >0.5 and < 1.0

    Only if the edge detector output is below the noise filter threshold will the filter be used

     Only if the edge detector output is above the edge filter threshold will the filter be used

     If the sensor is very high-end,turning High-Resolution Edge Detection on may get better results,but is is better to keep it off by default.

      Generally,for poor quality sensors with high noise levels,a smoothing filter should be used.For good quality sensors with low noise levels,a sharpening filter can be applied.

    Tuning过程中如何计算Fixed FPS MAX Line 和 Linear AFR Max Line的值?

    ov8825: preview size: 1632x1224

    Pixel per line X lines per frame = 3516x1264

    Max FPS: 30

    when you tune the exposure table, if you set fix fps, the fps should be fixed as 30,

    so the fixed FPS max line should be lines per frame in your driver code(1264). 

    And the AFR max line is decided that the max FPS you want to decrease in low light,

    if you want to keep FPS >=7.5 in low light, the max line should be 1264x30/7.5

    adb shell setprop persist.camera.stats.debug.mask 0 //disable adb shell getprop persist.camera.stats.debug.mask

    STATS_DEBUG_MASK_AEC_LOG (1<<0)   STATS_DEBUG_MASK_AWB_LOG (1<<1) STATS_DEBUG_MASK_AF_LOG  (1<<2) STATS_DEBUG_MASK_ASD_LOG (1<<3) STATS_DEBUG_MASK_AFD_LOG (1<<4)

    adb shell setprop persist.camera.tinless enable adb shell getprop persist.camera.tinless

    adb logcat | grep -i "simple-grey-world" adb logcat | grep -i "CCT:" adb logcat | grep -i "rg,bg,cnt" adb logcat | grep -i "prev_exp_index" adb logcat | grep -i "green-p" adb logcat | grep -i "final-awb-decision"

  • 相关阅读:
    LibSVM文本分类之工程中调用LibSVM进行文本分类
    交叉验证–Cross validation
    SkySeraph博主的GLCM特征学习
    Libsvm分类步骤
    VC技巧
    转载 libsvm vc 移植 实现多类分类
    FANN学习2之建立简单工程
    开源库FANN学习笔记1
    忙了一上午终于把形状特征搞定了啊
    终于看到结果了
  • 原文地址:https://www.cnblogs.com/whw19818/p/5917179.html
Copyright © 2020-2023  润新知