Trustest

TRUSTEST


TRUSTEST

CAMERA IQ TEST SPEC


VER 1.01





Trustest Co., LTD
Joy Chao / Sales Manager
02-8668-9993
0961-535-539
Autotest8888@gmail.com
Line:a0961535539





Revision History
Version Date (Y/M/D) Notes Writer
1.01 2020/07/06 Initial released May Su




　 　 　 　
　 　 　 　
　 　 　 　













APPROVALS
PREPARED BY CHECKED BY APPROVED BY
Simon Hu Edison Chou Jack Jun


目錄 Index

01 Camera Module Description P.04-P.08
02 SFR P.08-P.11
03 Relative Illumination Test P.11-P.14
04 Optical Center P.14-P.17
05 Color Ratio & Color Shading Tests P.18-P.20
06 Defective Pixel (bright mode) P.21-P.22
07 Defective Row and Column (bright mode) P.23-P.23
08 Blemish---method1 (Optional) P.24-P.27
09 Blemish---method2 (Optional) P.28-P.29
10 Defective Pixel (dark mode) P.29-P.32
11 Defective Row and column (dark mode) P.32-P.32
12 Dark shading P.33-P.33
13 Dark Level P.33-P.34
14 Appendix A: Bilinear Demosaic P.35-P.36
15 Appendix B: PDAF pixel correction P.37-P.38












01.Camera Module Description










Detail Camera
The Detail camera comprises an autofocus mechanism, lens barrel subassembly, IR cut filter, sensor cover, connector, FPC, EMI film, and copper foil.

The following table describes key design details of the Detail camera.


No. Component Item Specification
1 Sensor Vendor & PN
2 Sensor Optical format
3 Sensor Active array size
4 Sensor Captured image size
5 Sensor Frame rate
6 Sensor Pixel size
7 Sensor Chip size
8 Sensor CRA on sensor
9 Lens Vendor & PN
10 Lens Number of lens element
11 Lens Focal length




12 f/#
13 Optical distortion (f-tan)
14 Relative illumination
15 FOV (Diagonal)
16 Operation Range
17 Nyquist frequency
18 IR cut filter Thickness
19 VCM Vendor & PN
21 VCM Exterior size
22 VCM Lens holder ID to fit lens
barrel
23 VCM Focusing type
24 VCM Operating temperature
25 VCM driver
+ Vendor & PN
26 VCM driver
+ Serial control Interface
27 VCM driver
+ Supply Voltage (min / max)
28 VCM driver
+ VCM current consumption
29 VCM driver
+ Operating temperature
30 Power
supply AVDD (min / typ / max)
31 Power
supply DVDD (min / typ / max)
32 Power
supply DOVDD (min / typ / max)
33 Power
supply Total current




34 Module Serial data interface (sensor)
35 Module Serial control interface
(sensor)
36 Module Input clock frequency
(sensor)
37 Module Output formats (sensor)
38 Module Serial control interface
(EEPROM)
39 Module Serial control interface
(VCM)
40 Module Temperature range


Scene Camera


No. Component Item Specification
1 Sensor Vendor & PN
2 Sensor Optical format
3 Sensor Active array size
4 Sensor Captured image size
5 Sensor Frame rate
6 Sensor Pixel size



7 Chip size
8 CRA on sensor
9 Lens Vendor & PN
10 Lens Number of lens element
11 Lens Focal length
12 Lens f/#
13 Lens Optical distortion (f-tan)
14 Lens Relative illumination
15 Lens FOV (Diagonal)
16 Lens Operation Range
17 Lens Nyquist frequency
18 IR cut filter Thickness
19 Power
supply AVDD (min / typ / max)
20 Power
supply DVDD (min / typ / max)
21 Power
supply DOVDD (min / typ / max)
22 Module Serial control interface
(sensor)
23 Module Input clock frequency
(sensor)
24 Module Output formats (sensor)
25 Module Serial control interface
(EEPROM)
26 Module Temperature range




02.SFR
This test is to quantify sharpness performance using Spatial Frequency Resolution (SFR) score method.
Test Setup


Item Value Comment
Light source White LED panel Uniform illumination across FoV
Light color temperature 6500 +/-500 K
Light brightness 1500 +/-200 lux Check at LED panel
Test chart Chessboard chart
Test chart position
tolerance Test chart position
tolerance shall be
+/-0.5cm.
Image format Bayer RAW RAW10
Gamma correction OFF
Analog gain 1x Register: 0x204, Value: 0x00
Register: 0x205, Value: 0x00
Digital gain 1x Register: 0x20E, Value: 0x01
Register: 0x20F, Value: 0x00
White balance OFF
Exposure Auto 100x100 center AEC window to
achieve the image brightness target
Frame rate Per supplier setting 30fps suggested, Sony default value
Size of region of interest 40 x 40 ROI to calculate SFR
Image brightness target
for center ROI 800 +/- 40 No saturation at center ROI from Gr
and Gb channel using 10-bit (0- 1023DN)




Lens shading correction OFF
Sensor BLC ON Default sensor black level
compensation
Sensor DPC OFF Sensor defective pixel correction
Black level target 64 DN Subtract from 10-bit image
Frame average 1
Image covered area 100% FoV
Resolution 3280 x 2464


Image pre-processing
After image capture in RAW10 format:


●
●
●

●

perform bilinear demosaic
subtract black level target
convert the RGB image to the gray image Y based on the following equation: Y = 0.3R + 0.59G + 0.11B
Use the grayscale image for SFR calculations.

SFR measurement shall be calculated at spatial frequency 110lp/mm (Ny/4). If there is too much noise found in image, multiple frames may be averaged.
Measurements are made at 13 points in the image: center and each corner at 0.3, 0.6 and 0.8 field SFR is measured at both horizontal and vertical directions at each location.
SFR should be calculated in each ROI according to ISO 122333:




SFR corner-to-corner balance at vertical and horizontal direction are calculated from the 0.8F values at specified chart distance.
SFR corner-to-corner balance at vertical direction = 1 - SFRvertical min /SFRvertical max

SFR corner-to-corner balance at horizontal direction = 1 - SFR


horizontal min

/SFR


horizontal max


Documentation
Save all SFR values and SFR corner-to-corner imbalance values for future reference. In case of failure, save test image in RAW10 format for failure analysis.
Pass / Fail Criteria


Test P/F criteria
Test NFOV (Right) FATP inline WFOV (Left) FATP inline
SFR
@INF
@110lp/mm 0.0F >= 0.51

0.6F >=0.37
0.8F >=0.23 NA
SFR imbalance
0.8F@INF 0.8F <= 0.55 NA
SFR
@10cm (macro)
@110lp/mm 0.0F >= 0.48

0.6F >= 0.25
0.8F >= 0.14 NA
SFR imbalance
0.8F@10cm(macro) 0.8F <=0.65 NA
SFR NA 0.0F >= 0.41




@40 cm
@110lp/mm
0.6F >= 0.32
0.8F >= 0.24
SFR imbalance
0.8F@40 cm NA 0.8F <= 0.50


03.Relative Illumination Test

The relative illumination tests the fall-off of pixel intensity that is mainly the result of lens shading. Illumination measured at four corners of image are compared to the center region of interest.

Test setup


Item Value Comment
Light source White LED panel Uniform illumination across FoV
Light color temperature 5100 +/- 500 K
Light brightness 1000 +/- 200 lux Check at LED panel
Test chart Diffuse light panel
Test chart distance < 3cm Cover the entire FoV
<= 1cm distance preferred
Image format Bayer RAW RAW10
Gamma correction OFF
Analog gain 1x Register: 0x204, Value: 0x00
Register: 0x205, Value: 0x00
Digital gain 1x Register: 0x20E, Value: 0x01
Register: 0x20F, Value: 0x00
White balance OFF No AWB
Exposure Auto 100x100 center AEC window to
achieve the image brightness target
Frame rate Per supplier setting 30fps suggested




Size of region of interest 24 x 18 ROI to calculate RI
Image brightness target
for center ROI 800 +/- 40 No saturation at center ROI from Gr
and Gb channel using 10-bit (0- 1023DN)
Lens shading correction OFF
Sensor BLC ON Default sensor black level
compensation
Sensor DPC OFF Sensor defective pixel correction
Black level target 64 DN Subtract from 10-bit image
Frame average 1
Image covered area 100% FoV
Resolution 3280 x 2464


After image capture in RAW10 format:


●
●
●

●

perform bilinear demosaic
subtract black level target
convert the RGB image to the gray image Y based on the following equation: Y = 0.3R + 0.59G + 0.11B
Use the grayscale image for calculations.


Overlay five ROI onto the image. In each ROI, calculate the average pixel values. The RI and RI_balance of Y channel is calculated as follows:

RI = min (ROIcorner

) / ROI


center

RI_balance = min(ROIcorner) / max(ROIcorner )

Addtionally,the average intensity of each ROI (Center, Upper Left, Upper Right, Lower Left, Lower Right) shall be recorded .




Pass/Fail Criteria


Test item NFOV (Right) FATP WFOV (Left) FATP
RI >= 0.25 (25%) >= 0.27
RI_balance >= 0.80 (80%) >= 0.80
RI_roiCenter_mean
RI_roiBR_mean
RI_roiBL_mean
RI_roiTR_mean
RI_roiTL_mean Record value only Record value only


Documentation
Supplier shall save all corner RI values and RI_balance value for future reference. In case of failure, supplier shall save test image in RAW10 format for failure analysis.

04.Optical Center
Optical center is calculated from image taken in relative illumination test. OC shift quantifies how much off of the center bright region from sensor center.
光學中心 Optical Center 演算概念
1. 決定選取的ROI範圍 (可全取或部分擷取)











2. 二值化將同一亮度的輪廓取出







3. 計算該輪廓質心位置, 即視為Optical Center
X-Optical Center = ∫x G(x) dx ⁄ ∫G(x) dx .
Y-Optical Center = ∫yG(y) dy ⁄ ∫G(y) dy .

4. 取光學中心與影像中心的差 (dX, dY) 即為光學中心偏差度























5. Imatest Optical Center Method:



6. Trustest 方法 與 Imatest 方法差異:
Imatest 在步驟中提到會找最亮值之後再取該亮度的 0.95 倍以上的像素點進行積分計算, 與Trustest方法應該差異不大, 且Trustest方法更具有 Robust 特性
(例如中心與周圍亮度差異過小的狀況, 可能在 Imatest 方法中會有潛在風險)


Setup and Image Pre-processing
She grayscale image generated by the Relative Intensity Test shall be used for Optical Center calculation

Any of various center finding methods, such as centroid estimation or polynomial fit may be used.

Pass/Fail Criteria

Test item NFOV (Right) P/F criteria WFOV (Left) P/F Criteria
OC shift [μm] 75 70
OC shift [# pixels] Record value only Record value only
OC_center_X [pixel index] Record values only Record values only




OC_center_Y [pixel index]
OC_shift_X [# pixels]
OC_shift_Y [# pixels] Record values only Record values only



























05.Color Ratio & Color Shading Tests
Color ratio and shading tests are to catch defects from the sensor, lens, or coverglass drive that could deteriorate the color performance of the camera

Test setup
Tse same capture conditions as Relative Illumination Test (above)

Image Pre-processing:

After image capture in RAW10 format:


●
●
●
●

●

subtract black level target
separate data into individual channels (1640 x 1232)
Perform PDAF compensation on Red Channel (see appendix for details) center each channel image to 1632 x 1224 --- i.e. remove first 4 rows, last 4 rows, first 4 columns, and last 4 columns of data.
Divide the image into 16x12 blocks - there will be 102 x 102 block

In each channel, group the center 2x2 blocks as one entity and calculate it’s average value.



Color Ratio:
Using the center 2x2 block ROI average from each color channel, record three ratios: R/Gr, B/Gr, Gb/Gr

Color Shading:

For the color shading test, in each channel, calculate the average value for each 16x12 ROI. The Ratio of channels in each ROI is compared to the ratio of their center 2x2 block average. Calculate the ratio for each 16x12 ROI. Report the maximum ratio for △ R/Gr and △ B/Gb


△ R/Gr = |(R/Gr) - (R/Gr)center


| / (R/Gr)


center

where, (R/Gr) = ratio of averaged pixel value of R and Gr of a block under test
(R/Gr)center = ratio of averaged pixel value of center 2 x 2 blocks R and Gr

△ B/Gb = |(B/Gb) - (B/Gb)


center

| / (B/Gb)center

where, (B/Gb) = ratio of averaged pixel value of B and Gb of a block under test

(B/Gb)


center

= ratio of averaged pixel value of center 2 x 2 blocks B and

Gb
Pass / Fail Values


Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
Color_R_Gr_Ratio 0.4 - 0.65 0.4 - 0.65
Color_B_Gr_Ratio 0.4 - 0.65 0.4 - 0.65
Color_Gb_Gr_Ratio 0.92 - 1.08 0.92 - 1.08
Color Shading
Max_Delta_R_Gr <= 0.40 <= 0.40
Color Shading
Max_Delta_B_Gb <= 0.40 <= 0.40


Documentation
Supplier shall save color ratio & shading values for future reference. In case of failure, supplier shall save test image in RAW10 format for failure analysis.


































06.Defective Pixel (bright mode)




This test is to look for defective pixel that does not show high pixel value by comparing it to its surrounding pixels.

Test Setup

Test setup is same as for Relative Illumination (above).

Image Pre-processing

After capturing image in RAW10 format,
● Separate image into R, Gr, Gb, B channels.
● Perform PDAF compensation on Red Channel (see appendix).

In each color channel (R, Gr, Gb, B), calculate average pixel value Cx of a ROI of 31 x 31 with pixel under test (PUT) located at the center of this ROI. This can be done, for example, with a 2D convolution of an averaging kernel. Then, go through all pixels and check how much PUT is off from the average.
R = 100% * |Vx-Cx| / Cx
where, Vx = pixel value of a PUT in given ROI in each channel (x = R, Gr, Gb, B)
Cx = averaged pixel value of a ROI in each channel (x = R, Gr, Gb, B)

The PUT is considered defective if R > 30%. The reported metric is a count of defective pixels for that channel.

Pass / Fail Values
Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
Defect_white_R_Singlet 0 - 65 0 - 65
Defect_white_Gr_Singlet 0 - 65 0 - 65
Defect_white_Gb_Singlet 0 - 65 0 - 65
Defect_white_B_Singlet 0 - 65 0 - 65

Documentation
In case of failure, supplier shall save test image in RAW10 format for failure analysis.



07.Defective Row and Column (bright mode)

Test Setup

Test setup is same as for Relative Illumination (above).
Image Pre-processing
After capturing image in RAW10 format, perform bilinear demosaic, then convert RGB image to the gray image Y based on the following equation:

Y = 0.3R + 0.59G + 0.11B

Averaged pixel value of a row under test is compared to the averaged pixel value of adjacent 8 rows above it and 8 rows below it. Similarly, averaged pixel value of a column under test is compared to the averaged pixel value of adjacent 8 columns of its left and 8 columns of its right. The edges are padded with mirror reflection.

If the difference is more than 2.5 DN, row or column under test is considered defective.
Pass / Fail Criteria
No rows or columns shall exceed 2.5 DN difference.

Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
defect_bright_row 0 0
defect _bright_col 0 0

Documentation
In case of failure, supplier shall save test image in RAW10 format for failure analysis.






08.Blemish---method1 (Optional)
The goal of blemish test is to catch any low-contrast and large area defects such as foreign particle or assembly damage in the optical path.

Test Setup
Test setup is same as for Relative Illumination (above).
Image Pre-processing
Image pre-processing is more intensive for Blemish test than other tests. Each pixel is queried with an ROI-based analysis. To minimize edge effects of this analysis, the
starting image is padded on all sides and then a simple lens shading removal is performed

Full details:
After capturing image in RAW10 format,


●
●
●
●

●

●


Perform PDAF pixel correctioon
perform bilinear demosaic
subtract black level target (64 DN)
convert the RGB image to the gray image Y based on the following equation: Y = 0.3R + 0.59G + 0.11B
Perform padding on all edges. The pad size is 89 pixels.

Starting with the example of padding the top side, create ROI_1, with the first 89 rows of the image, and ROI_2 with rows 90-178. Subtract ROI1 pixel-wise from ROI_2, and then take the average of each column to create a single row. Copy that row 89 times to make an 3280 x 89 array. Subtract this array from ROI_1 and then prepend the resulting array to the image --- which will now be 3280 x 2553.

Repeat this process at the bottom edge of the image and at each side (with 89- column) padding. The resulting image will be 3458 x 2642.

Perform lens shading correction to avoid false fails near edges.
○ Generate a 32 x 24 gain table by bicubic downsampling of the 2458 x 2642 image.
○ Normalize the gain table against its maximum value.
○ Scale the normalized gain table back to 3458 x 2642 using bicubic interpolation, to ensure smoothness.
○ Apply the normalized gain table to the padded image 3458 x 2642 image by element-wise multiplication.


The LSC-corrected Y image is now ready for uniformity calculation.

The uniformity calculation is performed on each of the pixels in the central 3280 x 2464 of the LSC-corrected image, the pixels corresponding to the original unpadded image.

Create a 3x3 array of 33x33 ROIs centered about the Pixel Under Test (PUT). Determine the average value of each ROI.

For each PUT (n,m) calculate a value



RUm,n = 1 −

𝑅𝑂 𝐼𝑚, max⁡(𝑅𝑂𝐼𝑛𝑒𝑖𝑔ℎ𝑏𝑜𝑟𝑠


)


This RU array should be the dimensions of the original image.


Next create an RU map, whose value is 1 where RU exceeds the threshold defined for that region. The RU map is divided into 4 regions: Corner, Edge, Outer, and Center, whose defect thresholds and locations are defined as:

Region Threshold Where applied
Center 0.0185 Central region 201
pixels from edge of
image
Outer 0.03 Pixels in frame that
starts 97 pixels from
edge and ends 200
pixels from edge
Edge 0.06 peripheral 96 rows/
columns (excluding
corners)
Corner 0.1 33x33 regions at each
far corner of image.

Finally, perform a continuity check is performed on the binary RU map. If there are 2x13 or 13x2 regions of continuous 1s in the RU map, the module fails the test.



Pass / Fail Criteria


Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
Blemish_count

Number of instances of
13x2 or 2x13 contiguous
regions above threshold 0 0
Spot_size

Largest continuous region
in the continuity check Record only Record only

09. Blemish ---method2 (Optional)
The goal of blemish test is to catch any low-contrast and large area defects such as foreign particle or assembly damage in the optical path.

Test Setup
Test setup is same as for Relative Illumination (above).
Image Pre-processing
Image pre-processing is more intensive for Blemish test than other tests.

After capturing image in RAW10 format,
• Top row padding is applied as a pre-process for lens shading correction to apply a smooth boundary condition to pixel data region.
•
1. After image capture, perform bilinear demosaic, subtract black level target, then convert the RGB image to the gray image Y based on the following equation.
Y = 0.3R + 0.59G + 0.11B
2. At each ROI size given in table below, calculate the average pixel value of a ROI under test (ROI m,n ) and compare it with its 8 surrounding region of
interest, 𝑅𝑂𝐼𝑛𝑒𝑖𝑔ℎ𝑏𝑜𝑟 with the following equation.
RUm,n = max[ |𝑅𝑂𝐼𝑛𝑒𝑖𝑔ℎ𝑏𝑜𝑟 − 𝑅𝑂𝐼𝑚,𝑛| ]




10.Defective Pixel (dark mode)

This test is similar to the bright mode defective pixel calculation, but is somewhat simpler because lens shading does not confound the result. A simple threshold can be used for evaluation, rather than using a relative % threshold as in the bright mode case).

Test Setup


Item Value Comment
Light source N/A Dark panel
Light color temperature N/A
Light brightness Completely dark Checked at top of camera module
Test chart N/A
Test chart distance N/A Use black box
Image format Bayer RAW RAW10
Color Interpolation OFF
Gamma correction OFF




Analog gain 1x per Sony test
spec A: 0x204, D: 0x00
A: 0x205, D: 0x00
Digital gain 1x per Sony test
spec A: 0x20E, D: 0x01
A: 0x20F, D: 0x00
White balance OFF No AWB
Exposure 66ms A: 0x340, D: 0x14
A: 0x341, D: 0x42
A: 0x342, D: 0x0E
A: 0x343, D: 0x58
A: 0x202, D: 0x14
A: 0x203, D: 0x38
Frame rate Per supplier setting Low fps suggested, e.g. 15fps
Size of region of interest 31 x 31
Image brightness target
for center ROI N/A
Lens shading correction OFF
Sensor Black Level
Compensation ON
Sensor DPC OFF Sensor defective pixel correction
Black level target 64 DN 10-bit
Frame average 1
Image covered area 100% FoV
Resolution 3280 x 2464


Image Pre-processing

After capturing image in RAW10 format,
● Separate image into R, Gr, Gb, B channels.
● Perform PDAF compensation on Red Channel (see appendix).

In each color channel (R, Gr, Gb, B), calculate average pixel value Cx of a ROI of 31 x 31 with pixel under test (PUT) located at the center of this ROI. This can be done, for example, with a 2D convolution of an averaging kernel. Compare PUT against averaged pixel value of ROI as follows:
R = |P-Q|
where, P = pixel value of PUT in ROI



Q = averaged pixel value of ROI
.
Pass / Fail Values
PUT is considered defective if R >= 28 DN. The reported metric is a count of defective pixels for that channel.
Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
Defect_dark_R_Singlet 0 - 975 0 - 975
Defect_dark_Gr_Singlet 0 - 975 0 - 975
Defect_dark_Gb_Singlet 0 - 975 0 - 975
Defect_dark_B_Singlet 0 - 975 0 - 975

Documentation
In case of failure, supplier shall save test image in RAW10 format for failure analysis.

11.Defective Row and column (dark mode)

Test Setup
Test setup is the same for dark mode defective pixel.
Image Pre-processing
Image is processed according to the bright mode defective row and column test. Pass / Fail Criteria
No rows or columns shall exceed 2.5 DN difference.

Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
defect_dark_row 0 0
defect _dark_col 0 0

Documentation
In case of failure, supplier shall save test image in RAW10 format for failure analysis.




12.Dark shading
The purpose of dark shading test is to find low frequency non-uniformity that domes from either sensor itself or light leakage through the assembly.

Test Setup
The test setup and image capture is the same as for dark mode defective pixel test. Image pre-processing

After capturing image in RAW10 format,
● Subtract black level target (64 DN) from image
● Separate image into R, Gr, Gb, B channels.
● Perform PDAF compensation on Red Channel (see appendix).

For each color plane, the image is divided into 33 x 33 ROIs. The last column and row of ROIs should contain the extra pixel remainder pixels left since the image size is not divisible by 33. For 3280 x 2464, the last ROI will be 46x55, the ROIs in the last row will be 46 x 33 and the ROIs in the last column should be 55x33.
The average of each ROI is calculated and the difference between the maximum and minimum ROIs is the dark shading value for that channel.

Pass/Fail Criteria
Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
Dark_Shading_R 0-10 0-10
Dark_shading_Gr 0-10 0-10
Dark_shading_Gb 0-10 0-10
Dark_Shading_B 0-10 0-10

Documentation
In case of failure, supplier shall save test image in RAW10 format for failure analysis and shall record the maximum and minimum pixel value.

13.Dark Level
Test Setup
The test setup and image capture is the same as for dark mode defective pixel test.



Image pre-processing

After capturing image in RAW10 format,
● Separate image into R, Gr, Gb, B channels.
● Perform PDAF compensation on Red Channel (see appendix).

The Dark Level value is simply the average value of all pixels in the channel. Pass/Fail

1. After image capture, separate it into R, Gr, Gb, and B channel
2. the averaged pixel value of the center 200x200 pixels ROI.
Criteria


Test item NFOV (Right) P/F criteria WFOV (Left) P/F criteria
Dark_Level_R 61-69 61-69
Dark_Level_Gr 61-69 61-69
Dark_Level_Gb 61-69 61-69
Dark_Level_B 61-69 61-69



14.Appendix A: Bilinear Demosaic
IMX355 has an RGGB mosaic.
Bilinear demosaic uses the following rules for an RGGB color filter array:

For Red pixels:
● R value: that pixel’s 10bit value.
● G value: average value of 4 nearest green pixels (up, down, left right), these are in a ‘+’ arrangement around the red pixel
● B value: average of 4 nearest blue values (upper left, upper right, lower right, lower left); these are in an ‘x’ arrangement around the red pixel.

For Gr pixels
● R value: horizontal average of the 2 red pixels to the right and left of the Gr pixel

●
●

G value: that pixel’s 10bit value
B value: vertical average of the 2 blue pixels above and below Gr pixel.


For Gb pixels
● R value: vertical average of the 2 red pixels above and below the Gb pixel ● G value: that pixel’s 10bit value
● B value: horizontal average of the 2 blue pixels to the left and right of the Gb pixel.

For Blue pixels:
● R value: average of 4 nearest red values (upper left, upper right, lower right, lower left); these are in an ‘x’ arrangement around the blue pixel.
● G value is the average value of 4 nearest green pixels (up, down, left right), these are in a ‘+’ arrangement around the red pixel

●

B value is that pixel ’s 10bit value.


At array boundaries, only the available pixels are averaged. See example below:




Pixel location Channel Pixel value
R11 R R11
R11 G Average (G12, G21)
R11 B B22
G12 R Average (R11, R13)
G12 G G12
G12 B B22
B22 R Average (R11, R13, R31,
R33)
B22 G Average (G12, G21, G23,
G32)
B22 B B22


In general, bilinear interpolation generates significant artifacts, especially across edges and other high-frequency content since it doesn‘t take into account the correlation among the RGB values.



15.Appendix B: PDAF pixel correction

PDAF layout
In IMX355, PDAF pixels are implemented in the Red Channel.

The following diagram illustrates Sony’s PDAF spacing, but has several errors. The spacing and indices within the 32x32 block illustrated on the left are accurate, however, the (0,0) pixel in each block is blue, such that the PDAF pixels fall on the Red Channel. The number of 32x32 blocks is correct, but the description of the perimeter padding is inaccurate. Accurate indices into the first block are shown below.


The PDAF pixels are implemented in blocks of 32 x 32 pixels.

If we start indexing at 0 and assign the upper left pixel in the sensor array (0,0) ((col, row)) and the lower right pixel (3279,2463), the first PDAF pixel is at index (40, 50), such that the origin pixel (0,0) of the 32x32 block is at (37,45).

In this new frame of reference of the first 32x 32 block, PDAF pixels occur at the following indices:
(3,5) (11,5) (19,5) (27,5)
(3,13) (11,13) (19,13) (27,13)
(7,21) (15,21) (23,21) (31,21)
(7,29) (15,29) (23, 29) (31,29)

PDAF pixels are in every 8th row. Column spacing is also 8, but the columns in the second 2 rows are shifted by 4 with respect to the first 2 rows.

The last PDAF pixel is at (3236, 2410).

There are 74 rows 32x32 blocks, and 100 columns.




0
1

0 1
37,45

7 1 1 1 2 2

29

2

5


1


2


2

Compensation:
To match scheme used by camera module vendor, assign to the PDAF pixel, the value of the red pixel to the left of the PDAF pixel. For example, for the first PDAF pixel at (40,50), use the value from pixel (38,50).