This section includes the outline of the suggested technique used for image compression with the support of CNN-RF model to execute the image compression efficiently with minimum PSNR. This section also discusses the details about the implementation plan. The subsection includes:
3.1 Proposed workflow flowchart
3.2 Hybrid CNN-RF algorithm In this proposed CNN-RF approach towards image compression, the initial step is the population size. Here, the pixels of image is the population.
Random Forest algorithm is an ensemble learning technique which could be applied for classification or regression analysis and it operates by developing multiple decision trees at the training phase. Then, the aggregation of results by majority vote factor for classification or average factor for regression. Moreover, the random selection handles complete data with numerous variables running into thousands.
In this research, CNN is deployed for grouping of pixels. In convolution neural network, only a trivial portion of input layer neurons associate to neuron hidden layer. Here, the pooling layer is mainly deployed to minimize the dimensionality of the feature map and there exist numerous multiple activation and pooling layers enclosed within hidden layer of CNN. The fully connected layer forms the previous few layers in the network where the input to the fully connected layer is the output from final pooling or the convolution layer which is compressed and then fed to the fully connected layer. In grouping, depending on the least pixel value, the pixels are grouped and checks for informative and non-informative data.
CNN is an advanced form of multilayer neural network which is mainly entailed of input, hidden and output layers. A neuron is basically an elementary form of information processing unit of a CNN which comprises of a system of synapses or links. Every connection is classified as weight W1, W2,..., Wm, an adder function (linear combiner Eq. (1)) which calculates the weighted sum of the inputs,
and activation function f() for controlling the amplitude of the output of the neuron. The typical model of the neuron could be observed as Eq. (2)
Where Xi(i=1,2,3,...,n) specifies the input vector. Wi signifies the weights among 2 connective neurons. θ is the threshold. f() is the activations function, the frequently used function is sigmoid function Eq. (3)
y is the desired output.
Once the grouping is over, then the image segmentation of pixels takes place in which the image is divided into various subgroups known as image segments that further reduces complexity of image. Following image compression which compresses the image without degrading the quality and essential features of the image. This allows for further images to be stored in a certain volume of disk or memory space.