Is this Interlacing normal on PS2 footage? (OSSC Pro)

[Gamewise]

Hi all,

I’m using the OSSC Pro to capture footage on the PS2 and i’ve encountered quite a lot of interlacing issues during movement. (Scaled mode) I’ve tried quite a lot of solutions but I’m struggling to reduce it. I’m currently on motion adaptive deinterlacing, but bob or weave seem to be slightly worse.

I’m unsure if this is just a normal side effect of upscaling or something that can be reduced.

The issue seems to be a bit better in 1080p.

I’m using pro RGB cable from my PS2. Hdmi out into an elgato 4k capture card. Video appears more interlaced on the elgato output than the display. The display is a 4k samsung UHD TV.

Hope you guys have some tips.
Thanks for the help!

• This topic was modified 3 weeks ago by Gamewise.
• This topic was modified 3 weeks ago by Gamewise.

[Zacabeb]

The adaptive de-interlacing is quite basic in its interpretation of the picture, so it does make mistakes. Altera has a more sophisticated version than is used in the OSSC Pro, but it consumes substantially more resources.

The de-interlacing logic measures the difference in contrast across fields between the same and neighboring pixels to determine if there is motion. The resulting value controls whether the pixel is left as-is (weaving) or is given a weighted mean of the value across fields (effectively bobbing or blending.) The Motion Shift adjustment controls the threshold between those modes by dividing the motion value by a power of 2 (shifting the bits in the variable right to decrease the value, hence the somewhat unintuitive name Motion Shift.) At 0 it basically forces bobbing and at 7 it forces weaving.

Dropping Motion Shift to 1 or 2 may reduce the artifacts, but at the cost of increased bobbing in static parts of the image.

The more sophisticated version Altera has uses Sobel edge detection to derive motion vectors and more intelligently determine what to do, but I don’t know if it’s realistic to implement in the OSSC Pro given the additional performance overhead. Sharp pixel graphics are also more unforgiving since they more easily confuse the de-interlacer and reveal artifacts more clearly.

[marqs]

The Sobel edge deinterlacer needs more logic resources than what is available in current FW, but making an alternative image which cuts some other functionality to enable fitting could be an option. A second downside of it is reported deinterlacing latency (1 field versus couple lines) which however doesn’t make sense as it shouldn’t really need it. Perhaps it’s just a typo in datasheet so it’s something which would need to be measured.

[Author]

Posts