[pratham-mcw]

Summary

• In cv::ximgproc::weightedMedianFilter(), filterCore() processes the image one column at a time.
• in a single thread, each column clears and rebuilds its own joint-histogram and BCB, so no state is carried from one column to the next.

Changes

• Moved the column loop into parallel_for_ so different columns can run on different threads.
• Each worker allocates its own joint-histogram (H), BCB and necklace-link buffers (Hf, Hb, BCBf, BCBb), so threads share no writable state. Each column writes only its own pixels of outImg.
• For images narrower than 256 columns, nstripes is set to 1 so the loop runs as a single stripe and avoids parallel_for_ overhead on small inputs.
• The per-column algorithm is unchanged, so the output is exactly the same as before.

Performance Benchmarks