Databases
StackOverflow2013
Postsdbo
PO
 

Note that there are some explanatory texts on larger screens.

plurals
  1. PO
    primarykey
    data
    text
    <p>You've asked a lot of questions and I don't think I can answer all of them, but here are answers to as much of your question as I can.</p> <ol> <li><p>This is most certainly a nearest-neighbor algorithm where the goal is to find the two closest points to each point in the first vector and then check whether the ratio of their distances is less than some cutoff value.</p></li> <li><p>You <em>could</em> do this with a quadtree or kd-tree, but because your points are all one-dimensional values you could do much better using a balanced binary search tree. Given such a tree, if you thread a linked list through the nodes, you can find the k nearest neighbors to some test point p by looking up the closest element p in the binary search tree, then traversing (k + 1) steps in each direction and taking the k closest points of what you find. This runs in time O(lg n + k), where n is the number of points and k is as above. This is substantially more efficient than what you have now, which takes O(n) time per lookup.</p> <p>If the dimensionality of your feature vector is more than 1, but less than 20, then using kd-trees would be a very effective measure.</p> <p>For higher dimensionalities, you might want to either reduce the number of dimensions with PCA before applying a kd-tree or use a more scalable ANN structure, such as locality-sensitive hashing.</p></li> <li><p>SURF is best for scene and object detection. If you need to find out if two images are the same, you would do better with global descriptor algorithms, such as GIST. The advantage of using a global descriptor is that you get a single vector for the whole image and image comparison is performed with simple Eucledian distance.</p></li> <li><p>You could definitely do this using MySQL because you don't need a kd-tree. A simple balanced binary tree should be sufficient.</p></li> <li><p>RANSAC is a method of estimating model parameters that is robust against outliers. It is useful for using SURF features for combining multiple photographs into a 3D scene.</p></li> <li><p>Checking for false positives is definitely a machine learning exercise and I'm not well-trained in that area. You could probably do this using a supervised learning algorithm (such as an SVM, boosted decision tree, or neural network), but I don't know enough to advise you on this.</p></li> </ol> <p>Hope this helps!</p>
    singulars
    1. This table or related slice is empty.
    1. POExplain this algorithm (Compare points in SURF algorithm)
      singulars
      1. PTQuestion
    1. PTAnswer
    1. USDon Reba
    1. UStemplatetypedef
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. POExplain this algorithm (Compare points in SURF algorithm)
      singulars
      1. PTQuestion
    1. This table or related slice is empty.
    1. VO
      singulars
      1. PO
      1. This table or related slice is empty.
      1. VTUpMod
    2. VO
      singulars
      1. PO
      1. This table or related slice is empty.
      1. VTUpMod
    3. VO
      singulars
      1. PO
      1. This table or related slice is empty.
      1. VTApproveEditSuggestion
    1. COsorry, i forgot to show you the point class, I think that I need 64/128 dimensions, so maybe for 1 million images I'll need a randomized kd-tree. I have been reading about GIST, now I'm using SURF and I have the local features, the concept of global features is new for me, i'll look for some more info. About MySQL if I use kd-tree I'll have to build the tree first. Thanks.
      singulars
      1. PO
      1. USWiliam
    2. CO@Wiliam: kd-trees perform best when the number of dimensions is below 20. For your problem, you might want to either reduce dimensionality first using something like PCA or use a more scalable ANN structure, such as locality-sensitive hashing.
      singulars
      1. PO
      1. USDon Reba
    3. CO@templatetypedef, uf, I don't know if that's the better solution. I have readed in some papers that they use randomized kd-tree to store the features, but I don't know more details...
      singulars
      1. PO
      1. USWiliam
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload