Deep Stacked Autoencoder-Based Automatic Emotion Recognition Using an Efficient Hybrid Local Texture Descriptor

Extracting an effective facial feature representation is the critical task for an automatic expression recognition system. Local Binary Pattern (LBP) is known to be a popular texture feature for facial expression recognition. However, only a few approaches utilize the relationship between local neighborhood pixels itself. This paper presents a Hybrid Local Texture Descriptor (HLTD) which is derived from the logical fusion of Local Neighborhood XNOR Patterns (LNXP) and LBP to investigate the potential of positional pixel relationship in automatic emotion recognition. The LNXP encodes texture information based on two nearest vertical and/or horizontal neighboring pixel of the current pixel whereas LBP encodes the center pixel relationship of the neighboring pixel. After logical feature fusion, the Deep Stacked Autoencoder (DSA) is established on the CK+, MMI and KDEF-dyn dataset and the results show that the proposed HLTD based approach outperforms many of the state of art methods with an average recognition rate of 97.5% for CK+, 94.1% for MMI and 88.5% for KDEF.

NPALOSS: NEIGHBORING PIXEL AFFINITY LOSS FOR SEMANTIC SEGMENTATION IN HIGH-RESOLUTION AERIAL IMAGERY

The performance of semantic segmentation in high-resolution aerial imagery has been improved rapidly through the introduction of deep fully convolutional neural network (FCN). However, due to the complexity of object shapes and sizes, the labeling accuracy of small-sized objects and object boundaries still need to be improved. In this paper, we propose a neighboring pixel affinity loss (NPALoss) to improve the segmentation performance of these hard pixels. Specifically, we address the issues of how to determine the classifying difficulty of one pixel and how to get the suitable weight margin between well-classified pixels and hard pixels. Firstly, we convert the first problem into a problem that the pixel categories in the neighborhood are the same or different. Based on this idea, we build a neighboring pixel affinity map by counting the pixel-pair relationships for each pixel in the search region. Secondly, we investigate different weight transformation strategies for the affinity map to explore the suitable weight margin and avoid gradient overflow. The logarithm compression strategy is better than the normalization strategy, especially the common logarithm. Finally, combining the affinity map and logarithm compression strategy, we build NPALoss to adaptively assign different weights for each pixel. Comparative experiments are conducted on the ISPRS Vaihingen dataset and several commonly-used state-of-the-art networks. We demonstrate that our proposed approach can achieve promising results.

Denoising of MRI images using fast NLM

Denoising of image is a very crucial step which should retain fine details but should remove noise. Making the difference between noise and actual edge related data is very difficult. NLM filter helps to make a differentiation between image data and noise data. Its weight function decides the weightage of the neighboring pixel depending upon the similarity with the pixel to process. It helps to retain the edges and avoid it from smoothening. This paper discusses the implementation of NLM filter using hardware platform Spartan 6. After implementaion of this on FPGA, not only denoise the image but preseve edges and there is a tremendous saving in time compared to its matlab implementation. Denoised image performance is calculated using various objective metrics such as MSE, PSNR, SSIM, PFOM etc. FPGA implementation shows clearly the advntages over its matlab implementation.

Performance of Predictive Coders for Wireless Capsule Endoscopy Image Compression

Wireless capsule endoscopy is a medical diagnostic technique developed for the endoscopic examination of the small bowel. The encoder module is the core of the wireless capsule endoscopic system impacting on power and area requirement for the hardware implementation of the capsule. One of the remarkable features of the endoscopic image is that the neighboring pixels are highly correlated. Two predictive coding techniques are considered in this work exploiting the above fact. The first predictive coder i.e., DPCM coder is based on previous horizontal neighboring pixel, whereas the second predictive coder is based on adjacent horizontal and diagonal neighbors. The performance of the predictive coders is tested with 41 small bowel type endoscopic images available in the Gastrolab dataset. The results show that the average compression rate and peak signal to noise ratio attained by DPCM coder and newly tested predictive coder are 66.37 % & 73.03 % and 32.17 dB & 35.55 dB, respectively

Merged Local Neighborhood Difference Pattern for Facial Expression Recognition

Facial expression based emotion recognition is one of the popular research domains in the computer vision field. Many machine vision-based feature extraction methods are available to increase the accuracy of the Facial Expression Recognition (FER). In feature extraction, neighboring pixel values are manipulated in different ways to encode the texture information of muscle movements. However, defining the robust feature descriptor is still a challenging task to handle the external factors. This paper introduces the Merged Local Neighborhood Difference Pattern (MLNDP) to encode and merge the two-level of representation. At the first level, each pixel is encoded with respect to center pixel, and at the second level, encoding is carried out based on the relationship with the closest neighboring pixel. Finally, two levels of encodings are logically merged to retain only the texture that is positively encoded from the two levels. Further, the feature dimension is reduced using chi-square statistical test, and the final classification is carried out using multiclass SVM on two datasets namely, CK+ and MMI. The proposed descriptor compared against other local descriptors such as LDP, LTP, LDN, and LGP. Experimental results show that our proposed feature descriptor is outperformed other descriptors with 97.86% on CK+ dataset and 95.29% on MMI dataset. The classifier comparison confirms the results that the combination of MLNDP with multiclass SVM performs better than other combinations in terms of local descriptor and classifier.

High Density Impulse Noise Removal using Advanced Median Filter for Preserving Image Quality Metrics

In this work, a procedure to remove the high density salt and pepper noise from a corrupted image is developed and to compare the output image with the original image through the image quality metrics. As a common practice the corrupted pixels are replaced by the median of neighboring pixel values by considering a constant number of neighboring pixels. But in this proposed method the corrupted pixels are identified and are replaced by the median of the neighboring pixel values which are adjustable, to preserve and improve the image quality metrics. This method makes a comparison between the corrupted and uncorrupted pixels and performs the median filtering process only on the corrupted ones. In this work a 3x3, 5x5 and 7x7 square neighborhood are used. The output images are observed with low neighborhood as well as high neighborhood pixel values. The calculation of PSNR (Peak Signal to Noise Ratio) and MSE (Mean square error) value for each dimension with different percentages are considered for the comparative analysis

Implementation of Pixel Likeness Weighted Frame (Plwf) Filter Technique Based Digital Image Denoising for DSP Applications

Digital images are often corrupted by contaminated display and information quality noise. Images can be corrupted at any stage during which they are acquired and transmitted through the media. Image denoising is a basic function designed to eliminate noise from naturally corrupted images. This work proposes a fixed-point discrete wavelet transform (DWT) architecture that uses a nonlinearly modified pixel-like weighted frame (PLWF) technique to denoise the highthroughput of adaptive white Gaussian white noise (AWGN) images. The linearized state to be based on the neighboring pixel unity is that the state model noise is used to improve the peak signal to the sound rate (PSNR). The proposed architecture is employed in two different stages - consistent and conditional sorting output selection unit. The detailed result of the proposed architecture shows the size and display quality of any state-ofthe-art performance and some recently introduced work. For further evaluation of the denoising capability, the algorithm is compared to some state-of-the-art algorithms and experimental results on simulated sound images and captured images of lowlight noise especially large image processes Low noise light picked up by the test results. The performance of the proposed method is compared to wavelet thresholds, bilateral filters, nonlocal averaging filters, and bilateral multi-resolution filters. The study found that the draft production plan is smaller than the wavelet threshold, the bilateral filter, and the non-local means of filtering and larger superior/similar to the method, visual quality, PSNR and image index noise bilateral multi-resolution filter quality

A Novel super voxel-based 3D segmentation method for Irritable Bowel Syndrome

The main aim of this research paper is to implement a model-driven machine learning based adaptive 3D Segmentation Scheme for detecting the IBS (Irritable bowel syndrome) disease. This algorithm taking into account by endoscopy driven visual images for the purpose of machine analyzing and convert that 2D RGB coordinates into 3D RGB coordinates for improving the accuracy of the segmentation. In previous segmentation schemes, the IBS images are obtained by the use of ultrasound imaginary technique, but the main issue of the imaginary was the noise present in the images. We are overcoming this issue by applying the endoscopy images. Adaptive smoothing technique used in pre-processing stages with neighboring pixel reference. The feature data extraction stages estimate the shape and color and region-based features for segmentation. The proposed scheme performance with our 50 image Database shows that the results accuracy of proposed system outperforms multiple conventional segmentation methods

Image Segmentation on Satellite based Image Using Advanced Watershed Method

Picture division is the method toward unscrambling a portrait into numerous parts. This be regularly worn to distinguish substance or other considerable data in advanced pictures. Readily available are a wide range of approaches to perform picture division; including One of the keys in characterization is the division. Portioning a picture into districts is an issue that has numerous conceivable arrangements. Question based division has been extremely prominent as of late as a result of its remarkable capacity to isolate the adaptability and homogeneity concerning outline and shading starting its neighboring pixel cell particularly near the informational index among towering spatial inconstancy. In any case, the most basic is the decision of parameter esteems. This investigation expects to improve the division by picking suitable filtration technique and parameter esteems. Notwithstanding, to decide the execution of the division procedure, in this venture we perform assortment Fit catalogue metric utilizing ArcPy bundle. Here are 5 examination regions chose various state & scope. Grades demonstrate with the purpose of bigger territories give the most astounding precision in AFI assessment. Be that as it may, this is a differentiation to the grouping comes about which gives higher exactness towards the littler dataset

Content-based image retrieval using local texture features in distributed environment

In this paper, we propose novel content-based image retrieval (CBIR) algorithms using Local Octa Patterns (LOtP), Local Hexadeca Patterns (LHdP) and Direction Encoded Local Binary Pattern (DELBP). LOtP and LHdP encode the relationship between center pixel and its neighbors based on the pixels’ direction obtained by considering the horizontal, vertical and diagonal pixels for derivative calculations. In DELBP, direction of a referenced pixel is determined by considering every neighboring pixel for derivative calculations which results in 256 directions. For this resultant direction encoded image, we have obtained LBP which is considered as feature vector. The proposed method’s performance is compared to that of Local Tetra Patterns (LTrP) using benchmark image databases viz., Corel 1000 (DB1) and Brodatz textures (DB2). Performance analysis shows that LOtP improves the average precision from 59.31% to 64.36% on DB1, and from 83.24% to 85.95% on DB2, LHdP improves it to 65.82% on DB1 and to 87.49% on DB2 and DELBP improves it to 60.35% on DB1 and to 86.12% on DB2 as compared to that of LTrP. Also, DELBP reduces the feature vector length by 66.62% as compared to that of LTrP. To reduce the retrieval time, the proposed algorithms are implemented on a Hadoop cluster consisting of 116 nodes and tested using Corel 10K (DB3), Mirflickr 100,000 (DB4) and ImageNet 511,380 (DB5) databases.