nan Aprinaldi

Human Sperm tracking using Particle Swarm Optimization combined with Smoothing Stochastic sampling on low frame rate video

In this paper, we present a technique for visual tracking in the field of Human Sperm motion. Application of sperm cell tracking is mainly important in Intracytoplasmic Sperm Injection (ICSI), a medical procedure that has enabled the In Vitro Fertilization (IVF) of a single sperm which is injected directly into an egg. In this paper, we consider the problem of tracking single object in video sequences of human sperms and a newly developed Smoothing Stochastic Approximate Monte Carlo (SSAMC) based tracker enhanced by Particle Swarm Optimization (PSO). The problem for this research is that the motility or movement of Human Sperm is fast and unpredictable. In addition, each and every sperms have closely similar size and shape. To solve this problem, we used PSO for searching algorithm (finding the best target) in a Search Window, it can reduce the search space in every each consecutive frame. The measurement results of the proposed method are then compared with the manual measurements done by experts. The experiment results were conducted on both open video data and our own video data. Experiment results showed that the proposed method can handle our specific problem in human sperm cell tracking, and give us a better result as compared to our previous tracker, which used geometric transition dynamic model and without any enhancement by PSO.

Automatic fetal organs segmentation using multilayer super pixel and image moment feature

Segmentation of fetal organs such as head, femur, and humérus on ultrasound image is one of the challenges in realization of automated system for fetal biometry measurements. Although many methods have been developed to overcome this problem, most of them are generally specific to one organ of the body alone. The research in this paper will focus on a machine learning method that has been available before: multilayer super pixel classification using random forest. The focus of this study is to improve the accuracy by exploring compactness parameter in the formation of super-pixels. In addition, we also add moment image features such as translation, rotation, and scale invariant to improve the segmentation performance. The experimental results showed that the difference in compactness parameters will provide different result for the accuracy, Fl-score, recall, and specificity. The addition of moment features can also improve the performance of image segmentation of fetal organs even though increase was not significant. Fetal head segmentation using proposed method has higher Fl-score and specificity, but lower accuracy and recall compared to previous methods. Whereas fetal femur and humérus segmentation using proposed method has higher accuracy, Fl-score, recall and specificity compared to previous method.

ArcPSO: Ellipse detection method using particle swarm optimization and arc combination

In this paper we present a technique for ellipse detection in digital images based on swarm intelligence algorithm and arc segment combination. The proposed method is then used as embryo quality scoring assessment during the first 24-48 hours since its morphological structure can be approximated by ellipse. The idea of the proposed algorithm are based on combining possible arcs for the ellipse shaped objects and try to find the best combinations using Particle Swarm Optimization technique to find the actual ellipse. The process involves detecting line segments in the image and then followed by arc segment extraction from lines to get potential elliptical arcs. The detection process is then guided by Particle Swarm Optimization (PSO) by utilizing the calculation of the fitness function from the arc segment that had been detected previously. The measurement results of proposed method are then compared with manual measurements. The experiment results were conducted on both synthetic data and real embryo images. Experiment results showed that the proposed method is better than several ellipse detection methods such as RHT, IRHT, and PSORHT to detect ellipses on the image. Another advantage of our proposed algorithm compared to the Hough Transform variants is that it can be used for multiple ellipse detection.

ECG signal compression by predictive coding and Set Partitioning in Hierarchical Trees (SPIHT)

In this paper we present a method for multi-lead ECG signal compression using Predictive Coding combined with Set Partitioning In Hierarchical Trees (SPIHT). We utilize linear prediction between the beats to exploit the high correlation among those beats. This method can optimize the redundancy between adjacent samples and adjacent beats. Predictive coding is the next step after beat reordering step. The purpose of using predictive coding is to minimize amplitude variance of 2D ECG array so the compression error can be minimize. The experiments from selected records from MIT-BIH arrhythmia database shows that the proposed method is more efficient for ECG signal compression compared with original SPIHT and relatively have lower distortion with the same compression ratios compared to the other wavelet transformation techniques.

Prediction the number of blastomere in time-lapse embryo using Conditional Random Field (CRF) method based on Bag of Visual Words (BoVW)

In vitro fertilization technology is used to help couples get children. During the development of IVF, embryological will observe the process of cleavage embryo until it is determined which gives the highest probability to produce a pregnancy. During this division process, the observation is done manually by embryological which produce subjective assessment of an embryo and vulnerable reduced quality embryos. Embryo quality is reduced due to the observation carried out outside a developed embryo. In addition to the number of embryos that increasingly divide and have a morphology that are difficult to observe, these judgements are prone to error than the embryological own. This research proposed a method to predict the number of blastameres of the embryo time-lapse using Conditional Random Field (CRF) based on Bag of Visual Words (BoVW). BoVW approach is used to represent data with the purpose of solving the problem of subjectivity embryological votes. The data used for the experiment is the data of the human embryo from the hospital Cipto Mangun Kusumo (RSCM) and data from mouse embryos. Data embryos RSCM has more variations than data in mouse embryo. Based on the experimental results, the use of BoVW able to overcome the problem of subjectivity embryological votes with an accuracy of¿ 80%. Besides the experimental results with the proposed method uses data embryos RSCM which difficulty level is higher than the data from moose embryos, they were able to identify with both the average accuracy of 96.79%.

Ellipse detection on embryo image using modification of arc Particle Swarm Optimization (ArcPSO) based arc segment

In Vitro Fertilization (IVF) technology is used to help couple that have problem in their reproduction organs. However, the success rate of IVF is just 30%, so it is a challenging task to increase that success rate. In this paper, we proposed an ellipse detection method on single and multiple embryo image by modifying the ArcPSO method on ellipse fitting process and the process on extract the arc segment for ellipse detection. The experiment results on single embryo data showed that the proposed method has a 24% better on F-Measure, 21% better on precision, and 28% better on recall as compared to the original ArcPSO method. While the experiment results on multi embryo data showed that the proposed method has a 7% better on F-Measure, 11% better on Precision, and 2% better on Recall as compared to the original ArcPSO method.

Multi codebook LVQ-based artificial neural network using clustering approach

In this paper we proposed multicodebook LVQ-based artificial neural network classifier using clustering approach. The classifiers are LVQ, LVQ2-1, GLVQ, and FNGLVQ. The clustering algorithm used to build multi codebook is K-Means, IK-Means, and GMM. Experiment result shows that on synthteic dataset multi codebook FNGLVQ using GMM clustering has higest improvement with 19,53% mprovement compared to FNGLVQ. Whereas on bencmark dataset multi codebook LVQ2-1 using K-Means clustering has higest improvement with 5,83% improvement cmpared to LVQ-2.1.

Learning semantic segmentation score in weakly supervised convolutional neural network

Semantic segmentation is an image labeling process for each pixels according to defined objects class and its presence in an image. Labeling process consists of recognizing, detecting location and labeling pixels that defines the object in the image. Annotation result of semantic segmentation needs ground truth to verify accuracy of score prediction. Therefore, this research propose a model to predict score of annotation accuracy. By casting the problem into constraining object boundary recognition, we described the annotation using foreground mask. To extract the feature, we used convolution neural network. We only used CNN trained on a image level annotation. In order to be able to infer the pixel instance, we adapt CNN architecture into weakly supervised learning. Experiments were conducted by finetuning Convolution Neural Network for object recognition using weakly supervised architecture for multilabel classification. In this paper we proposed to score semantic segmentation based on bag level information without the availability of pixel level annotation.

An efficient implementation of generalized extreme studentized deviate (GESD) on field programmable gate array (FPGA)

Outliers are one of the data problem which cannot be overlooked. The problem arise from the possibility of the outliers to become a bad data which can lead to error. One of the technique for identifying as well as eliminating outliers is generalized extreme student deviate (GESD). In this research the implementation of GESD on FPGA is examined. There are two complicated operations on GESD, i.e. division and square root which would be evaluated to find the most efficient processor design. The combination of Binomial and CORDIC has the most efficient resource utilizations with 48%. Meanwhile, the combination of Binomial and Non Restoring is the runner up with 56%.

A NOVEL KNOWLEDGE-COMPATIBILITY BENCHMARKER FOR SEMANTIC SEGMENTATION

The quality of a semantic annotation is typically measured with its averaged class-accuracy value, whose computation requires scarce ground-truth annotations. We observe that humans accumulate knowledge through their vision and believe that the quality of a semantic annotation is proportionally related to its compatibility with the vision-based knowledge. We propose a knowledge-compatibility benchmarker, whose backbone is a regression machine. It takes as input a semantic annotation and the vision-based knowledge, then outputs an estimate of the corresponding averaged class-accuracy value. The knowledge encodes three kinds of information, namely: cooccurrence statistics, scene properties and relative positions. We introduce three types of feature vectors for regression. Each specifies the characteristics of a probability vector that captures the compatibility between an annotation and each kind of the knowledge. Experiment results show that the Gradient Boosting regression outperforms the ν-Support Vector regression. It achieves best performance at an R2-score of 0.737 and an MSE of 0.034. This indicates not only that the vision-based knowledge resembles humans’ common sense but also that the feature vector for regression is justifiable. Index-terms: vision-based knowledge, knowledge-compatibility benchmarker, semantic segmentation, averaged class accuracy, regression