Article Preview
TopIntroduction
In recent image processing methods, de-noising has a major part. De-noising is the method of eliminating noise from the image and preserving all the relevant features of the image. While de-noising the images, the noises get suppressed without affecting the features such as important particulars of edges or textures from the images. Medical images are subjected to noise corruption during transmitting and receiving process. Every analog and digital recorder has noise susceptibility. Medical images are normally affected because of device/detector specifications and environment conditions. Henceforth, lessening of noise remains the same for conventional problem in medical imaging. The foremost objective lies in efficient de-noising of the high noisy dense images, with no additional computational expenses. Gaussian noise may also include salt, pepper and speckle noises. The removal of these noises is necessary for effective feature selection process.
However, de-noising the medical images to get the high quality is the greatest problem these days. The cause for this problem in de-noising of medical images is not unique. So, for effective noise removal in medical images, denoising is performed by DTCWT. It is the improved form of discrete wavelet transform and it also contains other important properties (Fan, 2019). The DWT can generate the real and imaginary coefficients if the filters are precisely planned, diverse from other filters (Selesnick et al. 2005). Thus, complex-valued filtering is used in the DT-CWT. The complex signals are disintegrated as real coefficients in the transform domain. It is needed for precisely explaining the energy localization of oscillating performance. Alternate method for implementing an expansive CWT includes the application of Hilbert transform initially. Here, the input and Hilbert transformed data are implemented with real wavelet transform and the wavelet transform coefficients are merged to get CWT. In case of dimensional signal, the redundancy factor of DT-CWT has 2 dimensions and is much smaller than static DWT (Naimi et al. 2015).
The de-noised images undergo segmentation and then the features are extracted. Following this, feature selection is performed to get the best features of images. The feature selection process consists of two methods, wrapper and filter techniques (Zhu et al. 2007). Here, the wrapper technique utilizes the learning process for analyzing the feature subset. But, this wrapper technique is not suitable for high dimensional images because it requires large computation time. Thus, the filter technique is more suitable for the feature selection process since it does not require learning process and the computation time is less.
The objective of selecting the features from the dataset is to improve the functioning of the classifier. It improves the accuracy of the classifier and also provides better result. The classification process becomes faster with feature selection. If the number of features in the image is more, then the size of the search space also increases exponentially. However, it is difficult to use complete search methods to obtain the finest result. Also, the feature selection methods undergo immobility in local optima (Zawbaa et al. 2016). Thus, a hybrid moth flame optimization algorithm is proposed for feature set reduction. MFO algorithm is based on the movement of moth towards the natural light. It is known as transverse orientation. Also for improving the functioning of MFO, the opposition based learning technique is used. Tizhoosh (2006) has proposed the OBL technique which is then utilized for reinforcement learning acceleration. This OBL technique helps the MFO algorithm to obtain the most appropriate features.
This paper is arranged as: Section 2 discusses the methodological background of the proposed method. Section 3 details the proposed framework. The methodology of proposed work is explained in the Section 4. Section 5 details the experimental results. Data validation is given in Section 6. At last, Section 7 concludes the paper.