Find person you are mostly like based on HMM image recognition -- only for fun
2011-09-18 22:59
513 查看
At last, my experiment program for fun can run without crash.
These days, I am thinking to use HMM base image recognition to develop a tool which is used to seek a candidate's photo which best matches one selected target image, i.e. to see who matches the selected cartoon or celebrity's photo. Below is a snapshot
of this program so far. It is based on Open Cv 2.0.
![](http://hi.csdn.net/attachment/201109/18/0_1316357076jJil.gif)
In order to find amongst e.g. 20 input candidates' photos which matches the selected target one, at the begining, I thought using each candidate's photo to appy to this selected person's photo's HMM model calculating possibilities, but it seemed the images
with less contrastness always had higher possibilities. An extreme example is, a pure black or pure white image usually has the highest calculated possibility value, no matter which target photo it is matching against.
So, I changed the way of seeking later. First for each candidate's photo its most matched target is found, meaning that each candicate belonged to at most one target group. Program will only select the highest value amongst those which belongs to this target
group.
It is possible that there is no candidate found for one selected target since the selected target is not mostly matched by any candidate photo.
The following content is extracted from readme of this program:
==============================
This program is for experiencing HMM based image recognition,
based on OpenCv 2.0 library.
It is only for fun, to find which photo mostly matches the
target image, e.g. to find which person's photo is mostly like
the selected cartoon figure or a celebrity's photo.
Following are steps how this program works:
1) Go to menu "File" -> "Open", open a *.jpg or *.bmp picture which
is then shown in the right window. Then use the left button of
your mouse to select the area on this picture. Next, go to menu
"File" -> "Add Object", type in the name of this photo, then its
selected part appears in the left window. You can then repeat
this step 1) to add more photos for the same target person and
add photos for other persons.
2) After adding several cartoon or celebritie's images, go to menu "File"
-> "Train" to train these photos, namely, to generate HMM parameters
for each person in the left window. The HMM parameters for each person are
generated according to their own photos and are used later for measuring
which image matches them best.
3) Now, select a person in the left window, then click menu
"Seek The Lucky" to start finding a image which most matches
the selected one in the left window. In popped up dialog select
the candidates which are to be matched against the selected target
and amongst them the best one will be finally found and its name
is shown in a popped up dialog.
Note: All information could be saved into disk automatically. When you
open this program next time, go to meanu "File" -> "Open", select
"Celebrities File(*.txt)" and open the file named "untitled_base.txt"
which should be located at the same directory as your first input
image.
Due to time limitation, file name always be "untitled_base.txt", and
there is no other choice about which directory it can be saved at.
Last, it is claimed to only for fun, study and experiencing HMM image
recognition.
==========================
These days, I am thinking to use HMM base image recognition to develop a tool which is used to seek a candidate's photo which best matches one selected target image, i.e. to see who matches the selected cartoon or celebrity's photo. Below is a snapshot
of this program so far. It is based on Open Cv 2.0.
![](http://hi.csdn.net/attachment/201109/18/0_1316357076jJil.gif)
In order to find amongst e.g. 20 input candidates' photos which matches the selected target one, at the begining, I thought using each candidate's photo to appy to this selected person's photo's HMM model calculating possibilities, but it seemed the images
with less contrastness always had higher possibilities. An extreme example is, a pure black or pure white image usually has the highest calculated possibility value, no matter which target photo it is matching against.
So, I changed the way of seeking later. First for each candidate's photo its most matched target is found, meaning that each candicate belonged to at most one target group. Program will only select the highest value amongst those which belongs to this target
group.
It is possible that there is no candidate found for one selected target since the selected target is not mostly matched by any candidate photo.
The following content is extracted from readme of this program:
==============================
This program is for experiencing HMM based image recognition,
based on OpenCv 2.0 library.
It is only for fun, to find which photo mostly matches the
target image, e.g. to find which person's photo is mostly like
the selected cartoon figure or a celebrity's photo.
Following are steps how this program works:
1) Go to menu "File" -> "Open", open a *.jpg or *.bmp picture which
is then shown in the right window. Then use the left button of
your mouse to select the area on this picture. Next, go to menu
"File" -> "Add Object", type in the name of this photo, then its
selected part appears in the left window. You can then repeat
this step 1) to add more photos for the same target person and
add photos for other persons.
2) After adding several cartoon or celebritie's images, go to menu "File"
-> "Train" to train these photos, namely, to generate HMM parameters
for each person in the left window. The HMM parameters for each person are
generated according to their own photos and are used later for measuring
which image matches them best.
3) Now, select a person in the left window, then click menu
"Seek The Lucky" to start finding a image which most matches
the selected one in the left window. In popped up dialog select
the candidates which are to be matched against the selected target
and amongst them the best one will be finally found and its name
is shown in a popped up dialog.
Note: All information could be saved into disk automatically. When you
open this program next time, go to meanu "File" -> "Open", select
"Celebrities File(*.txt)" and open the file named "untitled_base.txt"
which should be located at the same directory as your first input
image.
Due to time limitation, file name always be "untitled_base.txt", and
there is no other choice about which directory it can be saved at.
Last, it is claimed to only for fun, study and experiencing HMM image
recognition.
==========================
相关文章推荐
- For those who are interested nVidia have made GPU Gems 1 available on their website. You can find it here
- You are attempting to build on a 32-bit system.Only 64-bit build environments are supported beyond froyo/2.2.
- Java SSH工程报错 Are you running on Java 1.4 or below with Apache Crimson? Upgrade to Apache Xerces (or Java 1.5) for full XSD suppo
- A survey on still image based human action recognition
- 论文阅读:BoVW-MI:TASK DRIVEN DICTIONARY LEARNING BASED ON MUTUAL INFORMATION FOR MEDICAL IMAGE CLASSIFIC
- Predicting Performance of a Face Recognition System Based on Image Quality
- Visual Attention Based on Long-Short Term Memory Model for Image Caption Generation 论文笔记
- You are attempting to build on a 32-bit system.Only 64-bit build environments are supported beyond froyo/2.2.
- What are you looking for on top of the world? English readings.
- You are attempting to build on a 32-bit system.Only 64-bit build environments are supported beyond f
- After logging in to the Lotus Notes Client, you are prompted for a password in DOS-like window (Viru
- You are attempting to build on a 32-bit system Only 64-bit build environments are supported beyond froyo/2.2
- build/core/main.mk:77: You are attempting to build on a 32-bit system. Only 64-bit build environment
- Image set based collaborative representation for face recognition
- 2013_CVPR_BoF meets HOG Feature Extraction based on Histograms of Oriented p.d.f Gradients for Image
- An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to S
- You are attempting to build on a 32-bit system.Only 64-bit build environments are supported beyond froyo/2.2.
- 论文笔记:An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application
- BaiXiang——【arXi2015】An End-to-End Trainable Neural Network for Image-based Sequence Recognition and
- Could not find the Visual SourceSafe Internet Web Service connection information for the specified database Would you like to launch the Visual sourceSafe connection wizard?