[swscaler @ 0x81ac360]No accelerated colorspace conversion found from yuv420p to bgr24

Surprisingly, there are only few people have such a problem after I googled for a while, and no solution as well.

I then traced the code and found that the issue should belong to FFmpeg.

The official FFmpeg library seems to be released without libswscale support in Ubuntu…

Anyway, the warning message can be removed by recompiling the FFmpeg with specifying
--enable-swscale, and according to the instruction:

Also note that --enable-gpl must be assigned since there is a condition in the libswscale/swscale.c source code:

The function yuv2rgb_get_func_ptr is in charge of the yuv420 to bgr24 colorspace conversion I think.

05/01/2011

1. you should recompile opencv to get the new FFmpeg work.
2. configure "--enable-shared" and "--enable-pic" in each library if you want to use shared libraries.

The traditional approach of computer graphics has been used to create a geometric model in 3D and try to reproject it onto a two-dimensional image.

Computer vision, conversely, is mostly focused on detecting, grouping, and extracting features (edges, faces, etc.) present in a given picture and then trying to interpret them as three-dimensional clues.

So I am more like a vision guy…

With VC++ 2010, compiling in debug mode is fine but the runtime kept crashing with some mysterious error code. The weird thing is with release mode both compiling and runtime is fine. Then I tried to compile OpenCV 2.1 by myself using VC++ 2010 (by CMake), the compiling process was OK, which gave me a number of *.lib and *.dll files, but it resulted in both modes runtime error.

It might work with MinGW since with ubuntu 10.10 it is fairly smooth.  Let’s see if OpenCV 2.2.0 would fix the issue.

Currently I would focus on Linux version first.

到底圖論裡面的碎形可以做什麼用途，一直是我心中的一個疑惑。直到接觸了關於高維資料的資料庫索引(indexing high-dimenional data)演算法，才發現除了碎形可以自身遞回產生神奇的美麗圖案外，一些特定的碎形規則還有一個特性，即是空間填滿(space-filling)。然而，當直接拿碎形來做索引應用時，會遇到許多根本與實作上的困難，前者包括邊界效應(boundary effect)、群集特性(clustering property)等；後者則有資料稀疏(sparsity)、大數(big number)等問題。這些都是我在實作時碰上的難題，其中邊界效應有許多研究特別針對此現象來做改善，這方面我相信已有不錯的解決方案，然其它議題仍有討論的空間。

如果上述這些問題都能克服的話，我相信用碎形所產生的空間填滿曲線(space-filling curve)，是一個很理想的高維資料索引演算法，尤其是當用其作近似最近鄰居搜尋(approximate neariest neibors search)。相較於常見的kd-tree、R-tree、Quad-tree等空間分割演算法，空間填滿曲線是直接將高維資料投影(mapping)進一維空間，於是我們即可利用所有針對一維資料索引所設計的資料結構(例如B+tree)，來完成高維資料的檢索！當然這是很理想的狀況，降維(dimensionality reduction)的過程中，會造成不可避免的資訊流失(information losing)，例如資料原來在高維空間時的拓撲(topology)關係，當投影進低維空間時遭到破壞，因此造成在低維度時的檢索錯誤，上述的邊界效應即是因此而產生。所以相關的前人研究裡，都儘可能地針對降維後所造成的破壞來做彌補，以提高檢索時的準確率。

以下從Wikipedia那邊貼幾個常見的curve，其中Z-curve是最簡單的例子。

於是二維資料到 mapping 進一維空間時就可以用以下的方式來索引：

三維資料的例子，四維以上就無法visualize了:

以下是另外兩個常見的space-filling curve，Hilbert curve有動畫可以示意：

Sierpiński curve:

Suddenly found that I have almost not blogged for one year…