فشرده سازي تصوير ويدئوئي براساس تلفيق روشها ي تطبيقي و شبكه اي

چكيده انگليسي :

Compression of Video Sequences Using Combination of Mesh based and Block Matching Methods Majid Mohrehkesh m mohrekesh@yahoo com Date of Submission September 8 2008 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor Dr Shadrokh Samavi samavi96@cc iut ac irAbstractVideo sequences require large memory space for storage and wide bandwidth for transmission Differentmethods and standards have been used for the mentioned applications Two main strategies for this purpose areblock matching algorithms BMA and mesh based methods Each of these strategies has its own strengths andweaknesses In block matching techniques it is assumed that pixels of block move to a new location withoutlosing their integrity as block Therefore block matching algorithms are not capable of dealing with rotations zoom ins and zoom outs But BMA s produce small errors and require smaller bandwidths In mesh basedalgorithms a frame is partitioned into for instance non overlapped right angle triangles Then by using themotion vectors of the three vertices of the triangle we can reconstruct the displaced triangle By doing so thezoom in or zoom out situations can be realized Also rotations can be modeled This gives the mesh basedmethods the advantage of distinguishing complex movements and reconstructing better quality images with noblocking effects But on the down side these methods produce larger errors In this thesis an algorithm is presented which adaptively uses the two mentioned strategies Initially auniform triangular mesh is superimposed on the current frame of the sequence Then the locations of the nodesare identified on the frame A block is placed around each node with the node being at center of the block Themotion vector of each node is found using the observed motion of its corresponding block When two objects arepassing each other in a frame a part of the mesh and its triangles are located at the boundaries of the two objects These boundary elements can cause large errors and require special care A voting mechanism can identify theboundaries If two of the three motion vectors are going in a specific direction and the third vector has a differentpath then boundary can be declared On the other hand if all three motion vectors of the three vertices have thesame general direction then it reveals a movement that is similar to a block movement Under these conditionswe can move the triangle using a translation This situation occurs when we have a stationary background andthe mesh triangle is located in the background or it can occur when the triangle is located inside a moving object A third situation occurs when each of the vertices is moving in a different direction This condition is anindication of one of the three situations If all three vertices are pointing towards a single point it means thatthere is zoom in situation If all motion vectors are pointing away from a point which is most likely locatedinside of the triangle then it can be concluded that the triangle is being zoomed out The third situation is whenmotion vectors are showing a spin around scenario This is an indication of a rotational movement in that part ofthe frame Hence voting is performed among the three parts of each triangle to identify objects that are passingin front of each other Subsequently translation interpolation or affine transformations is performed Theproposed method as compared to the mesh based method has higher PSNRs and in comparison with blockmatching algorithms produces better visual qualities The visual qualities of the reconstructed images beforeadding the estimation error to them are especially very high when there are moving objects passing each other This usually indicates that smaller compensation errors are required to reconstruct the original frames Furthermore smaller magnitudes of compensation errors indicate their localization around zero This in turn produces smaller

مهره كش، مجيد

فشرده سازي تصوير ويدئوئي براساس تلفيق روشها ي تطبيقي و شبكه اي

تخمين حركت , تبديلaffine , تطبيق بلوك , سنجش بندي