Video Segmentation decomposes a video into many frames throughout the sequence. Video segmentation applications are used in the field of robotics, video surveillance, traffic monitoring, video indexing etc. A group of pixels follow a analogous motion segmentation. Video is a series of frames ( pictures) displayed sequentially at fixed rate. All the pictures( frames) in a video files have equal size. Video contains continues series of 25 frames per second. And all the processing techniques are applied to frames.
Video segmentation technique accepts video as an input and the processed output will be an data extraction from input video or a new video. Video segmentation is a technique used for detecting changing frame in video. Video segmentation is classified into following types: Shape based video segmentation, edge based video segmentation, color based video segmentation and texture based video segmentation. This work is based on shape based video segmentation.
3. Related work
4. PROPOSED G-SEGON TECHNIQUE FOR OBJECT SEGMENTATION
Video segmentation is considered as an essential issue in the image coding filed. A large number of previous methods try to solve the segmentation problem from a certain perspective, e.g., threshold , template matching , region growing , edge detection and clustering. These methods have been proven to be successful in many applications, but none of them are generally applicable to all images and moving objects. This section, describe how the proposed method is used to segment the object from video. Proposed method G-SEGON provides segmentation more accurately than the existing method. First the background region is roughly detected by using opening and closing reconstruction then the mesh was constructed over the region. After constructing mesh , convert the frame into gradient and K-mean clustering algorithm is used to segment object by removing the background (BG). After segmentation compare both gradient segmented image and gray scale segmented image. Then the combined result is refined to get object. The block diagram of the proposed technique is shown in Figure.1
Pre-processing is necessary to remove the noises present in the image and to get accurate segmentation of object. The image cannot be applied directly for performing proposed technique due to the presence of noise in the image. To make it suitable for further operation pre-processing is done. It involves several processes such as RGB to grey level conversion, Image adjustment, and Region expansion.
‘ RGB to grey level conversion: In this step of pre-processing the input image or frame Ii is first converted into Grey level image .The image consists of grey shades, based upon the intensities the variation takes place from black to white. For the conversion of RGB to grey level representation, first step is to obtain the red, green, blue values. Then add the percentage of red, green, blue values.
‘ Image adjustment: In this step, quality of the image is improved by using image adjustment. Image adjustment is used for image enhancement.
‘ After image adjustment, region expansion is performed to get accurate position of object in the image. This result is used further process.
4.2 Morphological reconstruction operation
After pre-processing, we use multi-scale morphological operations to extract the features in the image Ii Dual multi-scale reconstruction operation was performed to segment object in the image. The opening reconstruction operation is performed until the original shape is recovered. To extract different object in the image particular structural element is used. This operation involves both opening and closing operation.
Opening: In the Opening operation the intensity of bright regions in the image is decreased. This is based on the structural element size compared to the bright regions. It involves both the erosion and dilation process. The opening operation is performed in two steps.
1. First the erosion of the image is taken
2. Then the result obtained by erosion undergoes dilation.
For opening operation consider a binary image Ibin with structural element in which the opening operation is performed by taking the erosion of the image Ibin then the result obtained undergoes dilation with structural element , which is given as:
Where, Ibin is the binary image.
is the structural element
Closing: Closing operation is reverse of opening operation. The Intensity of the background remained unchanged while processing the dark features. Closing operation takes place in two steps.
1. First the dilation of the image is taken.
2. Then the result obtained by dilation undergoes erosion.
For closing operation consider a binary image Ibin with structural element in which the opening operation is performed by taking the dilation of the image Ibin then the result obtained undergoes erosion with structural element , which is given as:
Where, Ibin is the binary image.
is the structural element
4.2 Back ground grey level variation
The technique that used in the object identification is extended to segment the object and their grey level by use both open and close reconstruction. To get stable outcome the reconstructed OR (CR) operation is not iterated completely so that the convex (concave) grey level variation of the image can be located. The object boundaries can be detected by using proper structural element. To obtain the stable grey-level variation morphological opening and closing operation is done simultaneously. This finds whether the concave and convex grey level variation is located or not. While performing OR (CR) operation the image false segmentation may occurs to avoid this smooth structural element operation is done across the boundary. For different morphological operation resultant image obtained is shown in the figure 3. Background grey level variation is identified by subtracting the image obtained from close operation with the open reconstruction operation.
IBG = ICR (i,j) – IOR (i,j)
Where IBG – is the Background variation of image.
ICR , IOR – is the image obtained by open and close operation.
– is the set of pixels.
In the figure 3 the image obtained by using various levels of morphological operation a) original image Ii ,b)closing reconstruction, c) Opening reconstruction, d) Closing-opening reconstruction operation.
4.3 Object region segmentation phase
In this step, the background grey level variation is extended over the image with the help of structural elements that used in the OR (CR) operation. Based on the OR (CR) operation the object in the image can be segmented by following three steps.
1) Initialization of object region:
Binary image mask Imask is the initial process of image segmentation. Binary Mask partitions the grey level image into object and background regions using top hat or bottom hat operation. It provides an outline of the object in the image. The binary mask of the image is shown in the figure 4.
Top hat operation:
Top hat operation helps to identify the brightest region present in the images by subtracting the original image from the resultant image of opening operation.
‘ Bottom hat operation uses the principal of removing the object that doesn’t fits the structural element used in the opening and closing operation.
‘ For this the original image is subtracted from the resultant image obtained from closing operation. This provides the image with only removed object.
‘ The region which has same grey-level variation between the bottom and top is considered as backgrounds.
2) BG grey level variational mesh
In this step, a background grey level mesh was constructed across the boundary region between the object and background in the image. When the background and the object regions are closely present in the image then we can’t get the information currently so we construct the variational mesh across the boundry. The grey level variational mesh was constructed by making use of isolated data points in the image by using Lagrangian interpolant algorithm.
Imesh = int erp(IBG)
After constructed mesh, image segmentation process is done through one efficient way the, k-means clustering is utilized to segment the object for gradient and grey scale images.
3) Gradient image segmentation using k-means clustering
In this step we use both gradient and k-mean operation to segment object from the image. Here, gradient operation is performed on Imesh and original image Ii to extract visual information. A gradient magnitude operator detects the amplitude edges at which pixel change their gray levels suddenly. The mathematical representation of gradient image is given as
is the gradient direction in the direction.
is the gradient direction in the direction.
Subsequently, IGD gradient image is obtained by taking the gradient of the input image and then subtracting it with gradient of the BG grey level variational mesh.
IGD = || Imesh(GD) ‘ Ii(GD) ||
IGD is the result gradient image
Imesh(GD) is the gradient of the BG grey level variational mesh
Ii(GD) is the gradient of the original image
After calculate IGD, the object region segmentation process is performed using k-means clustering. In image segmentation application, the observations are based on the pixels in the image plane. Consequently, K-mean clustering is applied for and then segmented is obtained.
The K-means segmentation algorithm is follows:
Step 1: Place K points into the space represented by the objects that are being clustered. These points represent initial group centroids.
Step 2: Assign each object to the group that has the closest centroid by following objective function
is a chosen distance measure between a data point and the cluster centre
Step 3: When all objects have been assigned, recalculate the positions of the K centroids.
Step 4: Repeat Steps 2 and 3 until the centroids no longer move. This produces a separation of the objects into groups from which the metric to be minimized can be calculated.
4) Gray level image segmentation using k means clustering:
On other hand, K-mean segmentation is applied directly on . The resultant grey scale image is obtained by subtracting the input grey scale image with the background (BG) grey level mesh.
IGL = || Imesh(GL) ‘ Ii(GL) ||
IGL is the result grey level image
Imesh(GL) is the grey level image of the BG grey level variational mesh
Ii(GL) is the original grey level image
4.4 Majority selection and Refinement phase:
After object region segmentation and , the combined mask ICM is done by majority selection process. In this step the segmented image obtained from both the grey level and the gradient operation are combined to get accurate segmentation of object in the image. For this we use majority selection procedure which is carried out by following steps,If both the pixel has same value i.e. 0 or 1 then no change is needed. For different value, the majority value in the neighbourhood pixels of particular pixel is taken and the pixel value is replaced by the majority value. The fig 5 shows the majority selection procedure. This procedure is repeated until the object is fully recovered from the image.
2. Compare both segmented gradient image and Grey scale image
3. Consider every pixel HC in both images
5. No change
7. Replace the pixel with majority of the neighbourhood pixel.
Fig 5 describes the majority selection technique for a) same pixel b) different pixel
Refinement of object and boundary region:
After performing the majority selection the object and boundary region in the image was refined. To avoid poor matching between the obtained image and the original image, coherent region justification procedure was done and the final object was extracted from the image I0. The overall G-SEGON procedure is given in the figure 6.
Where , I
Fig.6. Flowchart for G-SEGON procedure.
Figure 1: Process diagram of video segmentation