Vector Quantization and LBG Algorithm
Here, VQ has been experimented at three phases: encoder, channel, and decoder. The architecture of VQ is depicted in Fig. 13.1. The structure is constrained with three blocks where every block is comprised of diverse operating strategies. Block 1 is named as the encoder section. Image vectors can be produced by dividing an input image into adverse as well as nonoverlapping blocks. The origination of an effective codebook is the major operation in VQ. A codebook is composed of a collection of codewords of a size similar to a nonoverlapping block size. This model is named as an optimized one when the produced codebook is effective. Once a productive codebook is generated, every vector undergoes indexing with the help of the index value acquired from the index table. Block 2 is a channel by which indexed values can be forwarded to a receiver. Block 3 is a decoder section that encloses an index table, codebook, and reformed image. Hence, received indexed values undergo decoding with the receiver index table. The codebook that exists at the receiver is similar to the transmitter codebook.
The frequently used VQ in the LBG-VQ model is Generalized Lloyd Algorithm (GLA) termed as LBG approach. The distortion is reduced after implementing the LBG technique and offers a local codebook. The following steps are given below and also shown in Fig. 13.2:
- 1. Load with initial codebook Cj of size N. Assume the iteration value as m = 1 and initial distortion D{ — oo,
- 2. By applying codebook Cm = {YJ, partition the training set as cluster sets R, with the application of the nearest neighbor condition.
- 3. After the mapping of input vectors to initial code vectors is computed, determine the centroids of the partition area found in Step 2. It provides an enhanced codebook Cm+i-
- 4. Estimate the average distortion Dm+]. If Dm - Dm+, < T then terminates, else m = m + 1 and follow Step 2 to Step 4.
FIGURE 13.2 Overall process of the LBG algorithm.
Proposed MCS-LBG Algorithm–Based VQ
Fig. 13.3 shows the processes involved in the MCS-LBG algorithm for VQ. Initially, the original image is partitioned into nonoverlapping blocks. Then, the codebook generation takes place by the MCS-LBG algorithm. Then, the encoded data will be transmitted by the use of the indexing table. Then, the encoded data will be transmitted and is decoded by the use of the indexing table. Finally, the decoded data will be reconstructed by the use of the MCS-LBG algorithm.
CS Algorithm
The CSA that draws simulation from cuckoo’s change to breeding and replication is perfected by the suppositions as follows:
- • All cuckoos lay 1 egg in an arbitrarily chosen host nest at a time, as the egg signifies the feasible result to issue in hypothesis;
- • The CSA follows the existence of the fittest rule. Only the fittest between every host nest through the maximum amount of eggs are accepted on to the next generation;
- • The amount of host nests in the CSA is suitable early. A host bird spots the intruder solution through the possibility p„ e [0,1]. For such conditions, the host bird is either removed from the parasitic solution or abandons the nest entirely to seek a novel site to recreate the nest.
The steps involved in CSA are elitism, intensification, and diversification. Initially, a population of feasible results, a novel and potentially optimal solution (cuckoo egg) is created. When the fitness value of this cuckoo egg is high and arbitrarily choose the solutions from the obtainable host nests, it can be exchanged with the worst egg.
A cuckoo lays an egg at an arbitrary position using Levy flight that is characterized with
where Xj is the feasible egg, iter indicates the present generation number, Г is the
f °°
gamma function that is determined with the integral: T(x) = e~'tx~* dt, and X is a
constant (1 < X < 3). *^°
A Levy flight method is an arbitrary walk that procedures a range of instantaneous jumps selected from a heavy-tailed possibility density function. The step size u controls the arbitrary explore structures, which exploit the search space about the current best egg. It gets continued till search the exploration space more systematically with the distant domain randomization. So, the value for u should be allocated judiciously. A explore procedure is minimum efficient when u is selected as a tiny value, because the location to the novel created egg is close to the previous. Conversely, when the value for u is also large, the novel cuckoo solution can be located. For balancing the efficiency to both intensifications as well as diversification, the value of a is considered to be 1.
It is also indicated that the CSA outcomes of the GA as well as PSO with respect to the number of parameters to be changed. During CSA, only the possibility of the abandoned nests pa is changed. But, the setting of pa = 0.25 is necessary sufficient because it initiates out that the convergence amount of CSA is insensitive to pa. So, the fraction of nests to desert pa is considered as 0.25.
