Which hu moments must I used ??

[mangood]

Dear

I try to find Hu moment of my object in  image

I find in labview help that their are seven Hu moments as shown

which one is invariant. to scale and shift and rotate ???

best regards

[Bob_Schor]

This is not a LabVIEW question, but rather a question about image processing.  A Web seearch on "Hu moments" gives several references that may help answer your question (they seem to be designed to be invariant under various conditions).

[alexderjuengere · Accepted solution]

hi mangood,

as far as I remember Hu Moments 1 - 7 are centered and normalized, which makes them invariant to shift and scale.

Combining particular moments - which are invariant to scale and shift - are said to be invariant to rotation e. g. the Hu moment set.

So, all of them should be invariant to rotation

In terms of pattern recognition, I would use all of the 7 Hu Moments e.g. a Hu Moment vector.

But you should keep in mind, Hu Moments were desgined to distinguish between different typed characters.

Regards,
Alex

[mangood]

---

@Bob_Schor wrote:

This is not a LabVIEW question, but rather a question about image processing.  A Web seearch on "Hu moments" gives several references that may help answer your question (they seem to be designed to be invariant under various conditions).

---

dear its

it labview qesstion due to fact that the help of labview say that Hu moments invarient and i will use labview to find it

[mangood]

---

@alexderjuengere wrote:

hi mangood,

as far as I remember Hu Moments 1 - 7 are centered and normalized, which makes them invariant to shift and scale.

Combining particular moments - which are invariant to scale and shift - are said to be invariant to rotation e. g. the Hu moment set.

So, all of them should be invariant to rotation

In terms of pattern recognition, I would use all of the 7 Hu Moments e.g. a Hu Moment vector.

But you should keep in mind, Hu Moments were desgined to distinguish between different typed characters.

Regards,
Alex

---

---

@alexderjuengere wrote:

hi mangood,

as far as I remember Hu Moments 1 - 7 are centered and normalized, which makes them invariant to shift and scale.

Combining particular moments - which are invariant to scale and shift - are said to be invariant to rotation e. g. the Hu moment set.

So, all of them should be invariant to rotation

In terms of pattern recognition, I would use all of the 7 Hu Moments e.g. a Hu Moment vector.

But you should keep in mind, Hu Moments were desgined to distinguish between different typed characters.

Regards,
Alex

---

typed characters.???

is it global or for characters only???

[alexderjuengere]

typed characters.???

is it global or for characters only???

Hu moments are based on commonly accepted and respected math

Hu himself successfully tested his moments using various typed characters (It's been a while since I've read the article but I think those characters also were of the same font):

"It is shown that recognition of geometrical patterns and alphabetical characters independently of position, size and orientation can be accomplished.
It is also indicated that generalization is possible to include invariance with parallel projection"

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=1057692&searchWithin%3Dp_Authors%3A.Q...

From my experience with Hu Moments (programmed in LabView without any toolkits like Vision etc.) they can produce good results on other shapes besids typed characters, if your segmentation algorithm is producing objects in a reproducible manner.

Regards,
Alex