OBJECT    CLASSIFICATION
    IN FAR FIELD AND LOW
     RESOLUTION VIDEOS


1
             Proposed by    : Pr. S. Larabi
             Presented by   : Setitra Insaf
             Degree           : 2nd year doctoral student
             Universiy      : USTHB Algeria
             Email            : setita.insaf@hotmail.com
PLAN
   What is object classification
       Object classification can be either supervsed or
        unsupervised
   What is an object classfier
     Defintion
     Conditions
     types

   What are steps of object classification in videos
       Two approches : image based and background
        substraction
 Why is this work particular
                                                           2
 Conclusion
WHAT IS OBJECT CLASSIFICATION?
OBJECT CLASSIFICATION CAN BE EITHER SUPERVISED OR UNSUPERVISED




      Class label= car   Class label= person   Class label= bicyle …



                                                                       3
WHAT IS OBJECT CLASSIFICATION?
OBJECT CLASSIFICATION CAN BE EITHER SUPERVISED OR UNSUPERVISED


   Supervised classification= there is a supervisor for
    classifying:
       ~machine learning: machine learns from exemples
         First select exemples= near field/far field, indoor/outdoor, =/=
          scales and translation ..etc.
         Label all exemples

         Train the classifier with these exemples

         Apply the classification algorithm on new images




   Unsupervised learning: no supervisor for classifying
       ~Segmentation              Consider only supervised                  4
                                     classification for this
                                         presentatiion
WHAT IS AN OBJECT CLASSIFIER?

   Definition
       Algorithm of classification
   Conditions
       For robust classification:
         Works in different conditions (weather, resolution, compression
          …etc)
         Answer to a multiclass problem

         Works on realtime constrains

   Types
     SVM
     NN
     Naive Baysien
     …                                                                     5
What are steps of object classification in videos ?
    Two approches : image based and background substraction

 Image based: classification of every image alone
  and make a decision based on labels of every
  image
 Background Substraction(BS): use movment as a
  useful information for detecting regions of interest
  (moving objets) and then classify blobs resulting on
  BS




                                                              6
What are steps of object classification in videos ?
        Two approches : image based and background substraction

                                          BS classificication
   Image based
                                             Background
    classification                             substraction (mixture of
     Feature                                  gaussian, optical
                                               flow..etc)=>binary
      extraction/selection                     images with blobs
     Object classification                  Exemple of BS:

      (SVM, NN, Bayes                        Object tracking

      classifier…)                           Image improvement
                                               (removing supurious
                                               objects : shadows,
                                               false moving objects..)
                                             Feature
                                               extraction/selection
                                               (SIFT, Haar, HOG,
                                               DHOG ..)
                                             Object classification       7
Why is this work particular?


 Because pattern recognition is a wide area of
  research
 Because classifying in far field videos is a big
  challenge
     Small objects
     Occlusion
     Dificulties of application of existing algorithms (part based
      classifiers for exemple)
     Insatisfying results as far



                                                                      8
Conclusion

   Presentation of principal ideas of the work
   Seeking for people working on:
       Pattern recognition
       Clasification algorithms: SVM, NN, Bayes… (definitly, mathematical
        and probabilistic formulas are hard to understand!!!!)
       Feature extraction and selection
       Image and video retrieval
       …


          Thanks for attention dear collegues!!


              Questions? Feel free to ask !
                                                                             9

Object classification in far field and low resolution videos

  • 1.
    OBJECT CLASSIFICATION IN FAR FIELD AND LOW RESOLUTION VIDEOS 1 Proposed by : Pr. S. Larabi Presented by : Setitra Insaf Degree : 2nd year doctoral student Universiy : USTHB Algeria Email : setita.insaf@hotmail.com
  • 2.
    PLAN  What is object classification  Object classification can be either supervsed or unsupervised  What is an object classfier  Defintion  Conditions  types  What are steps of object classification in videos  Two approches : image based and background substraction  Why is this work particular 2  Conclusion
  • 3.
    WHAT IS OBJECTCLASSIFICATION? OBJECT CLASSIFICATION CAN BE EITHER SUPERVISED OR UNSUPERVISED Class label= car Class label= person Class label= bicyle … 3
  • 4.
    WHAT IS OBJECTCLASSIFICATION? OBJECT CLASSIFICATION CAN BE EITHER SUPERVISED OR UNSUPERVISED  Supervised classification= there is a supervisor for classifying:  ~machine learning: machine learns from exemples  First select exemples= near field/far field, indoor/outdoor, =/= scales and translation ..etc.  Label all exemples  Train the classifier with these exemples  Apply the classification algorithm on new images  Unsupervised learning: no supervisor for classifying  ~Segmentation Consider only supervised 4 classification for this presentatiion
  • 5.
    WHAT IS ANOBJECT CLASSIFIER?  Definition  Algorithm of classification  Conditions  For robust classification:  Works in different conditions (weather, resolution, compression …etc)  Answer to a multiclass problem  Works on realtime constrains  Types  SVM  NN  Naive Baysien  … 5
  • 6.
    What are stepsof object classification in videos ? Two approches : image based and background substraction  Image based: classification of every image alone and make a decision based on labels of every image  Background Substraction(BS): use movment as a useful information for detecting regions of interest (moving objets) and then classify blobs resulting on BS 6
  • 7.
    What are stepsof object classification in videos ? Two approches : image based and background substraction  BS classificication  Image based  Background classification substraction (mixture of  Feature gaussian, optical flow..etc)=>binary extraction/selection images with blobs  Object classification  Exemple of BS: (SVM, NN, Bayes  Object tracking classifier…)  Image improvement (removing supurious objects : shadows, false moving objects..)  Feature extraction/selection (SIFT, Haar, HOG, DHOG ..)  Object classification 7
  • 8.
    Why is thiswork particular?  Because pattern recognition is a wide area of research  Because classifying in far field videos is a big challenge  Small objects  Occlusion  Dificulties of application of existing algorithms (part based classifiers for exemple)  Insatisfying results as far 8
  • 9.
    Conclusion  Presentation of principal ideas of the work  Seeking for people working on:  Pattern recognition  Clasification algorithms: SVM, NN, Bayes… (definitly, mathematical and probabilistic formulas are hard to understand!!!!)  Feature extraction and selection  Image and video retrieval  … Thanks for attention dear collegues!! Questions? Feel free to ask ! 9