- Dr. R. R. Deshmukh, Professor, M.E., M.Sc. (CSE) Ph.D. FIETE, Presently working as Professor and Head and Program coo... moreDr. R. R. Deshmukh, Professor, M.E., M.Sc. (CSE) Ph.D. FIETE, Presently working as Professor and Head and Program coordinator for Department of Science and Technology sponsored FIST in Department of Computer Science and Information Technology, Dr. B.A.M. University, Aurangabad, (MS), India. He is a Fellow member and working as Chairman, IETE Aurangabad Centre, He has been elected as sectional member of ICT section of Indian Science congress Association. He is life member of ISCA, CSI, ISTE, IEEE, IAEng, CSTA, IDES and a senior member of ACEEE. He was Vice-Chairman of IETE Aurangabad Centre for last 4 years and member of Management Council of Dr. B. A. M. University. He is Member of Academic Council and Senate member of Dr. B. A. M. University, Aurangabad. Edited Nine books and published more than 95 research papers in reputed Journals, National and International conferences. He is reviewer and editor of several journals at nationaledit
Stock market data analysis needs the help of artificial intelligence and data mining techniques. The volatility of stock prices depends on gains or losses of certain companies. News articles are one of the most important factors which... more
Stock market data analysis needs the help of artificial intelligence and data mining techniques. The volatility of stock prices depends on gains or losses of certain companies. News articles are one of the most important factors which influence the stock market. This study basically shows the effect of emotion classification of financial news to the prediction of stock market prices. In order to find correlation between sentiment predicted from news and original stock price and to test efficient market hypothesis, we plot the sentiments of two companies (Infosys and Wipro) over a period of 10 years. For emotion classification, various classifiers such as Naive Bayes, Knn and SVM are evaluated. The comparison between positive sentiment curve and stock price trends reveals co-relation between them.
Research Interests:
The paper presents novel technique for recognizing faces. The proposed method uses mutual information for feature extraction techniques. Feed Forward Neural Network (FFNN) and Self Organizing Map Neural network (SOM) are used for... more
The paper presents novel technique for recognizing faces. The proposed method uses mutual information for feature extraction techniques. Feed Forward Neural Network (FFNN) and Self Organizing Map Neural network (SOM) are used for classification. Performance analysis is done by computing False Acceptance Ratio (FAR) and False Rejection Ratio (FRR). Experimentation is done over FACE4 database and achieved better performance.
Speech has much capability as an interface between human and computer which comes under the Human Computer interaction (HCI). The major challenge has been the nature of voice is ever varying speech signal. The paper presents the... more
Speech has much capability as an interface between human and computer which comes under the Human Computer interaction (HCI). The major challenge has been the nature of voice is ever varying speech signal. The paper presents the development of the speech recognition system using Swahili speech database which was collected in three sets: digits, isolated words and sentences from both native and non native speakers of Swahili language. Different feature extraction techniques deployed in the system are: Linear Prediction Coding (LPC) and Mel-Frequency Coefficients (MFCC). We have used the 12 coefficient features from MFCC and 20 coefficients features from LPC. All these features extracted techniques are applied and tested for the own developed Swahili speech database. Recognition and verification were done using confusion matrix and Support Vector Machine (SVM) as a classifier for the classification purpose. LDA was tested for the entire dataset for the dimension reduction. LDA gave a ...
Research Interests:
In the process of iris recognition localization and segmentation are the primary preprocessing step that locate the iris and segments it from the remaining part of the image. The normalization and feature extraction are very important and... more
In the process of iris recognition localization and segmentation are the primary preprocessing step that locate the iris and segments it from the remaining part of the image. The normalization and feature extraction are very important and crucial stages which prepare the iris image to extract the required unique features and create the templates that can be compared to find the uniqueness among the two irises. Normalization allows transformation of iris region into a fixed size dimensions that can be compared. In short normalization produces constant dimension size images of every localized input iris image which are ready to further processing and comparison of such images is possible due to their nature of constant dimension size. In general the normalization process prepares the segmented image for feature extraction process. This paper discusses the enhanced normalization process based on Daugman's rubber sheet model and feature extraction is based cumulative sum based chang...
In this research paper study of water quality & water level monitoring using remote sensing & GIS technique and require the satellite data for the study is reviewed. For human being we required fresh water so it is important to... more
In this research paper study of water quality & water level monitoring using remote sensing & GIS technique and require the satellite data for the study is reviewed. For human being we required fresh water so it is important to monitor The different water quality parameter i.e. physical, chemical, biological parameter are also reviewed in this paper. The optical property of water can be easily access. The different satellite sensor also reviewed in this paper. To calculate water level which satellite sensor is require and which technique is used.
Research Interests:
Humans can express their feelings by speaking, dancing, writing, etc., but speech is considered to be the easiest form of communication. There are various techniques on conversion of speech which are discussed in this paper. Since ancient... more
Humans can express their feelings by speaking, dancing, writing, etc., but speech is considered to be the easiest form of communication. There are various techniques on conversion of speech which are discussed in this paper. Since ancient times people communicate with each other by the use of some specific language. In India, various languages are spoken but Devnagri is a language, which plays an important role since ancient Indian society. In today’s world some modern languages have based on Devnagri like Marathi, Hindi, Sanskrit, and some South Indian languages also but the research in this field is very rare. The conversion of speech into text can be beneficial for the people who faces problem while communicating in society. The main objective of this paper is to summarize and compare various methods used in various stages of speech to text conversion.
Data de-duplication is one of the essential data compression techniques for eliminating duplicate copies of repeating data, and it has been widely used in cloud storage to reduce the amount of storage space and save bandwidth. To protect... more
Data de-duplication is one of the essential data compression techniques for eliminating duplicate copies of repeating data, and it has been widely used in cloud storage to reduce the amount of storage space and save bandwidth. To protect the privacy of sensitive data while supporting de-duplication, the salt encryption technique has been proposed to encrypt the data before its outsourcing. To protect the data security in a better way, this paper makes the first attempt to formally address the problem of authorized data de-duplication. Different from traditional de-duplication systems, the derivative privileges of users are further considered in duplicate check besides the data itself. We also present various new de-duplication constructions which supports the authorized duplicate check in hybrid cloud architecture. Security analysis demonstrates that the scheme which we used is secure in terms of the definitions specified in the proposed security model. We enhance our system in secu...
Research Interests:
Research Interests:
Autism Spectrum Disorder (ASD) is a multifaceted neurodevelopmental condition. Atypical communication mostly occurs in tandem with ASD. We compared voice pitch of 16 Marathi children and adolescents with ASD of age of 7 to 18 with 27... more
Autism Spectrum Disorder (ASD) is a multifaceted neurodevelopmental condition. Atypical communication mostly occurs in tandem with ASD. We compared voice pitch of 16 Marathi children and adolescents with ASD of age of 7 to 18 with 27 Typically Developing (TD). Speech samples have been recorded and stored in .wav format with sampling frequency of 48000 Hz. For analysis we used PRAAT, a program for speech analysis, manipulation and synthesis. We divided the ASD and TD group into total 4 groups on basis of age and gender for comparison. We found that differences in voice pitch are present in these comparison groups, and male ASD group have more pitch variation than respective comparison groups. In future we look forward to include more ASD participants in study to increase the Marathi speech database for ASD.
This paper gives a comparison of two extracted features namely pitch and formants for emotion recognition from speech. The research shows that various features namely prosodic and spectral have been used for emotion recognition from... more
This paper gives a comparison of two extracted features namely pitch and formants for emotion recognition from speech. The research shows that various features namely prosodic and spectral have been used for emotion recognition from speech. The database used for recognition purpose was developed on Marathi language using 100 speakers. We have extracted features pitch and formants. Angry, stress, admiration, teasing and shocking have been recognized on the basis of features energy and formants. The classification technique used here is KNearest Neighbor (KNN). The result for formants was about 100% which is comparatively better than that of energy which was 80% of accuracy. Keywords-Database, Emotion recognition, Feature Extraction, Formants, KNN classification , Pitch, Speech signals.
Research Interests:
Various operations performed by waiters like starting from taking orders till delivery of food/menu to the customer, also billing by cashier made manually. Due to manual process and paperwork may cause time delay, ignorance of customer,... more
Various operations performed by waiters like starting from taking orders till delivery of food/menu to the customer, also billing by cashier made manually. Due to manual process and paperwork may cause time delay, ignorance of customer, errors in billing leads to dissatisfaction of customers. As in today’s digital era, customers expect high quality, smart services from restaurant. So to improve quality of service and to achieve customer satisfaction, we proposed improvised E-Menu Recommendation System. This system can build e-reputation of restaurant and customer community in live. All orders and expenses are stored in database and give statistics for expenses and profit. The proposed recommender system uses wireless technology and menu recommender to build improvised E-Menu Recommendation System for customer-centric service. Professional feels and environment are provided to the customers/delegates with additional information about food/menu by using interactive graphics. Outcomes ...
Research Interests:
This paper proposes a system of isolated digit recognition for Marathi language using HTK approach. The database used contains 800 utterances of 40 individuals. Among them 20 are female and 20 are male. For training the acoustic features... more
This paper proposes a system of isolated digit recognition for Marathi language using HTK approach. The database used contains 800 utterances of 40 individuals. Among them 20 are female and 20 are male. For training the acoustic features of the database powerful MFCC i.e. Mel frequency cepstral coefficients technique is used. We have used word level model to recognize the Marathi isolated digits. The result analysis of the system shows 99.75% recognition with 48.75% accuracy.
Research Interests:
India is moving towards the direction of digital economy. The initiative of National projects like Digital India, Smart Cities and National Broadband Network has the direct impact on governance, transparency and accountability. As there... more
India is moving towards the direction of digital economy. The initiative of National projects like Digital India, Smart Cities and National Broadband Network has the direct impact on governance, transparency and accountability. As there is increase in the use of ICT for development and better governance, this rapid change towards a digital environment equally has brought forward the challenges of cyber security. According to national statistics, for last few years, out of all cybercrimes were reported across the country, Maharashtra state tops the list than any other states in India, so it becomes important to predict and analyze cybercrime trends for the future. This paper highlights the importance of data mining and machine learning and use the linear regression model to predict future cybercrime trends with reference to Maharashtra state. The real dataset of cybercrime is collected from the government website of Maharashtra state. Then the linear regression model is trained on th...
Palmprint is important member of biometric family, different types of algorithms and system have been proposed and great success has been achieved in Palmprint research, most of the previous Palmprint recognition works use white light... more
Palmprint is important member of biometric family, different types of algorithms and system have been proposed and great success has been achieved in Palmprint research, most of the previous Palmprint recognition works use white light source of illumination, which does not highlights the more feature these problem is solved by spectral band of multispectral Palmprint. This paper present feature level image fusion of multispectral Palmprint images for that purpose Polytechnique Hongkong University database is used. Initially the images were subjected to some preprocessing operation like filtering. Wavelet theory is introduced to resolve the Palmprint features extraction problem, and for matching purpose distance matrix is calculated by using Euclidian distance. Wavelet- based image fusion method is used as fusion strategy in our schema we have done fusion of Approximation Coefficient of RGB, NIR images, got the fused image by applying the DWT technique, to reduce the high dimensional...
Research Interests:
The earth crust is made up of variety of minerals. These minerals are having very significant applications in our day today life. The various studies, characterizing physical, chemical, electrical, structural properties, have been carried... more
The earth crust is made up of variety of minerals. These minerals are having very significant applications in our day today life. The various studies, characterizing physical, chemical, electrical, structural properties, have been carried out on the Lonar crater for studying mineralogy, surface morphology and geology but has not been done by remote sensing technology. So, the proposed work focused on exploring the mineralogy at the Lonar crater by using high resolution hyperspectral imageries. The spectral reflectance of minerals was characterized by using FieldSpec4 spectroradiometer. The minerals at Lonar crater were explored by performing preprocessing and spectral analysis. The techniques used in the work are Spectral Angle Mapper and Spectral Feature Fitting. The results of the work marked the presence of pigeonite and augite at Lonar crater which indicates that this crater is the result of extrusive volcanic activity. Also, the presence of augite underneath basaltic igneous ro...
Research Interests:
In recent years, mobile ad-hoc networks have rooted their pillars for emergency communication owing to reasonable cost, diversity, and easiness of mobile devices. The mobile ad-hoc networks is a self-coordinated, distributed and... more
In recent years, mobile ad-hoc networks have rooted their pillars for emergency communication owing to reasonable cost, diversity, and easiness of mobile devices. The mobile ad-hoc networks is a self-coordinated, distributed and infrastructure-less network of mobiles nodes. These characteristics of MANET enhanced the applicability of MANET in the field of emergency communication such as military and police operations, flood control and fire disaster management, etc. In MANET, a broadcast storm causes network problems as there are redundant broadcasts and packet collisions. Classical broadcast methods have motivated on evading broadcast storms by preventing some rebroadcasts. The further problem is the link breakages induced by node instability and their power exhaustion. In this research, we propose an adaptive neighbor knowledge-based hybrid broadcasting method to address these network problems. This method refines the counter threshold based on neighbourhood, mobility and energy of the node and makes use of the refined thresholds to make the broadcasting decision. The proposed method perform best as compared to AMECBB and TCBB by decreasing delay, packet dropping, and routing overhead and energy consumption.
Research Interests:
Research Interests:
Research Interests:
This paper studies three feature extraction methods, Mel-Frequency Cepstral Coefficients (MFCC), Power-Normalized Cepstral Coefficients (PNCC), and Modified Group Delay Function (ModGDF) for the development of an Automated Speech... more
This paper studies three feature extraction methods, Mel-Frequency Cepstral Coefficients (MFCC), Power-Normalized Cepstral Coefficients (PNCC), and Modified Group Delay Function (ModGDF) for the development of an Automated Speech Recognition System (ASR) in Arabic. The Support Vector Machine (SVM) algorithm processed the obtained features. These feature extraction algorithms extract speech or voice characteristics and process the group delay functionality calculated straight from the voice signal. These algorithms were deployed to extract audio forms from Arabic speakers. PNCC provided the best recognition results in Arabic speech in comparison with the other methods. Simulation results showed that PNCC and ModGDF were more accurate than MFCC in Arabic speech recognition.
Research Interests:
Spectroscopy is a rapid, simple, non-destructive and analytical technique, which provides a good alternative that may be used to replace conventional methods of soil analysis. Soil iron oxides occur in almost all type‟s soils and they... more
Spectroscopy is a rapid, simple, non-destructive and analytical technique, which provides a good alternative that may be used to replace conventional methods of soil analysis. Soil iron oxides occur in almost all type‟s soils and they reflect different environmental conditions by the high variability of their mineralogy and concentration. Soil iron oxide, being an important pedogenic indicator of the soil, measurement of Iron Oxide content can be used as an index of soil fertility. Analytical Spectral Device (ASD) Field Spec 4 Spectroradiometer is used which has 350-2500 nm spectral wavelength range to estimate iron oxide content from the soil sample. The Vis-NIR reflectance spectroscopy requires less effort and it is quick innovation to predict the soil iron oxide content. For collecting the soil iron oxide content from spectral data we are utilizing PLSR which is statistical regression method. This paper states the work that is done on different soil types at different places to observe the iron oxide content in soil. KeywordsIron Oxide, Reflectance Spectroscopy, ASD field spec 4, Vis-NIR, PLSR. __________________________________________________*****_________________________________________________
Research Interests:
Research Interests:
ABSTRACT
Research Interests:
Past research in mathematics, acoustics, and speech technology have provided many methods for converting data that can be considered as information if interpreted correctly. In order to find some statistically relevant information from... more
Past research in mathematics, acoustics, and speech technology have provided many methods for converting data that can be considered as information if interpreted correctly. In order to find some statistically relevant information from data, it is important to have mechanisms for reducing the information of each segment in the audio signal into features. These features should describe each segment in such a characteristic way that other similar segments can be grouped together by comparing their features. Pre-processing of speech signals is considered a crucial step in the development of a robust and efficient speech or speaker recognition system. This paper deals with comparative analysis of MFCC and DWT feature extraction technique.
Research Interests:
Palmprint recognition has received a lot of research attention and many systems have been proposed, most of the previous works use the white light sources for illumination. Recognition accuracy and anti-spoof capability is limited.... more
Palmprint recognition has received a lot of research attention and many systems have been proposed, most of the previous works use the white light sources for illumination. Recognition accuracy and anti-spoof capability is limited. Multispectral palmprint is a good technique to address these issues; the multispectral palmprint sensors typically collect spectral information in a Red peaking 660nm, Green peaking at 525nm, blue 470nm and NIR peaking at 880nm, selected discrete spectral bands, but which spectral band gives more discriminative information for that multispectral palmprint database is not sufficient, this problem is solved by hyperspectral Palmprint, which collects spectral information 420nm~1100nm, which could provide more discriminative information as compare to multispectral palmprint recognition. Optimized band selection could not only reduce the cost of illumination sources, reduce the data storage amount and save the computational time and improve the recognition acc...
Research Interests:
In 2010, there were approximately 217,730 new cases of prostate adenocarcinoma (CaP), the most common malignancy in men, and 32,050 deaths due to CaP in the United States . Approximately 17% of these men undergo some form of radiotherapy... more
In 2010, there were approximately 217,730 new cases of prostate adenocarcinoma (CaP), the most common malignancy in men, and 32,050 deaths due to CaP in the United States . Approximately 17% of these men undergo some form of radiotherapy such as intensity-modulated radiation therapy. During prostate radiotherapy planning, the prostate gland needs to be delineated on CT. Prostate carcinoma, the second most common cause of cancer death among American men, is not invariably lethal. A heterogeneous disease, it ranges from asymptomatic to rapidly progressive systemic malignancy. The prevalence of prostate cancer is so high that it could be considered a normal age-related phenomenon. Conventional MRI of the prostate relies on morphologic changes within the prostate to define the presence and extent of cancer. Segmentation of the prostate gland in Magnetic Resonance (MR) images is an important task for image guided prostate cancer therapy. In this paper we propose the automatic segmentatio...
In this paper, we describe the methodology derived for offline handwritten Sanskrit character recognition. This paper will provide a way for researcher to develop a dataset and techniques for offline handwritten Sanskrit character... more
In this paper, we describe the methodology derived for offline handwritten Sanskrit character recognition. This paper will provide a way for researcher to develop a dataset and techniques for offline handwritten Sanskrit character recognition. This paper describes basics of dataset; challenges associated with character system and proposed techniques to recognize Sanskrit Compound Characters.
Research Interests:
The automatic recognition of speech means enabling a natural and easy mode of communication between human and machine. Speech processing has vast applications in voice dialing, telephone communication, call routing, domestic appliances... more
The automatic recognition of speech means enabling a natural and easy mode of communication between human and machine. Speech processing has vast applications in voice dialing, telephone communication, call routing, domestic appliances control, Speech to Text conversion, Text to Speech conversion, lip synchronization, automation systems etc. Here we have discussed some mostly used feature extraction techniques like Mel frequency Cepstral Co-efficient (MFCC), Linear Predictive Coding (LPC) Analysis, Dynamic Time Wrapping (DTW), Relative Spectra Processing (RASTA) and Zero Crossings with Peak Amplitudes (ZCPA).Some parameters like RASTA and MFCC considers the nature of speech while it extracts the features, while LPC predicts the future features based on previous features.
Research Interests:
This paper describes the approach followed for development of speech database of Marathi digits starting from Shunya (zero) up to Nau (nine). The following paper describes the step by step procedure followed for the development of the... more
This paper describes the approach followed for development of speech database of Marathi digits starting from Shunya (zero) up to Nau (nine). The following paper describes the step by step procedure followed for the development of the speech database. For the development of automatic speech recognition (ASR) it is necessary to have a speech databases and the recognition rate depends upon the quality of the used speech databases. I. INTRODUCTION Speech is the way communication between humans where human can share their information with each other. The researchers around the world are trying to develop new interface system for communication between human and computer. Speech is having the capability of being used as a mode of interaction between human and Computer. Estimated number of languages spoken around the world varies between 6,000 and 7,000. Language technologies can play a vital role in the natural interfaces for those who can't understand the particular language. The lan...
Research Interests:
The research in the domain of the language technologies for Indian languages is far behind than the languages of developed nation. The work for the Indo-Aryan language, i.e. Marathi is behind. Development of speech database is the basic... more
The research in the domain of the language technologies for Indian languages is far behind than the languages of developed nation. The work for the Indo-Aryan language, i.e. Marathi is behind. Development of speech database is the basic need for developing an automatic speech recognition system. The accuracy of speech recognition depends on the quality of the speech data collected and the quality of training set data. This paper describes the progress in the development of isolated words Speech database of Marathi language for agriculture purpose.
Research Interests:
this paper illustrates the use of radial facial curves on 3D meshes to mode facial deformation caused by expression, occlusion and variation in poses and to recognize faces despite large expression, in presence of occlusion and pose... more
this paper illustrates the use of radial facial curves on 3D meshes to mode facial deformation caused by expression, occlusion and variation in poses and to recognize faces despite large expression, in presence of occlusion and pose variations. Here we represent facial surface by indexed collection of radial geodesic curves on 3D face meshes emanating from nose tip to the boundary of mesh and compare the facial shapes by comparing shapes of their corresponding curves. We use elastic shape analysis for comparing shapes of facial curves because elastic matching seems natural for facial deformation and is robust to challenges such as large facial expressions (especially those with open mouths), large pose variations, missing parts, and partial occlusions due to glasses, hair, and so on. Our results match or improve upon the state-of-the-art methods on two prominent databases:,GavabDB, and Bosphorus, each posing a different type of challenges.
Different new vertical domains are coming everyday so running a broad-based ranking model is no longer desirable as the domain are different and building a separate model for each domain is also not beneficial because there much time... more
Different new vertical domains are coming everyday so running a broad-based ranking model is no longer desirable as the domain are different and building a separate model for each domain is also not beneficial because there much time required for labeling the data and training the samples. In this paper we are handling the above problem by regularization based algorithm called as ranking adaptation SVM (RA-SVM), the algorithm is used to adapt existing ranking model of broad-based search engine to new domain. Here performance is still guaranteed and times taken to label the data training the samples are reduced. The algorithms only requires prediction from existing ranking model and do not require internal structure of it. Adapted ranking model concentrate on specific domain to achieve better results which are relevant to the search, further it reduces the searching cost also as the most appropriate search results are shown. Single ranking model is not good for training the search en...
Research Interests:
There are a variety of temporal and spectral features that can be extracted from human speech. These features are related to the pitch, Mel Frequency Cepstral Coefficients (MFCCs) and Formants of speech, can be classified using various... more
There are a variety of temporal and spectral features that can be extracted from human speech. These features are related to the pitch, Mel Frequency Cepstral Coefficients (MFCCs) and Formants of speech, can be classified using various algorithms. This study explores statistical features i.e. MFCCs and these features were classified with the help of Linear Discriminant Analaysis (LDA). This article also describes a database of artificial emotional Marathi speech. The data samples were collected from 5 Marathi movies (Actors and Actress) simulated the emotions producing the Marathi utterances which could be used in everyday communication and are interpretable in all applied emotions. The speech samples were distinguished by the various situations from the movie. The data samples were categorized in 5 basic categories that are Happy, Sad, Anger, Afraid and Surprise.
Speech is most common mode of communication between human. Human are trying to develop systems which can understand and accept the command via speech. This paper gives an overview of continuous speech recognition systems developed in... more
Speech is most common mode of communication between human. Human are trying to develop systems which can understand and accept the command via speech. This paper gives an overview of continuous speech recognition systems developed in different languages. This paper would be helpful for the researchers to find the brief overview of continuous speech recognition systems developed in different languages around the world and the recognition rate achieved by these system.
Emotion recognition from speech is an important area in research that represents human-computer interaction. The main purpose of this paper is to present literature review of different features and techniques used for speech emotion... more
Emotion recognition from speech is an important area in research that represents human-computer interaction. The main purpose of this paper is to present literature review of different features and techniques used for speech emotion recognition. The survey represents the importance of choosing different classification model and features for speech emotion recognition. Speech emotion recognition databases are also reviewed in this paper for the purpose of identifying the number of speakers, language used and emotion classification till date.
Research Interests:
Research Interests:
Research Interests:
ABSTRACT 3D model watermarking is a process of inserting data (like text, image, 3D model, audio etc.) into 3D model. Watermark remains in the 3D model even after applying various attacks. The technique proposed here is based on... more
ABSTRACT 3D model watermarking is a process of inserting data (like text, image, 3D model, audio etc.) into 3D model. Watermark remains in the 3D model even after applying various attacks. The technique proposed here is based on optimization, multi-resolution representation in wavelet domain, and DCT. In this method, the optimized parameter and the watermark scrambled by Arnold Transform is embedded in the DCT mid band coefficients after applying 3 levels Haar transform on 3D model. Blind method is used to extract the watermark. The system is tested and analysed against various 3D models. Analysis shows that the optimized parameter used, increases the PSNR and correlation values of the watermarked models and watermark resp. Results show that the system is robust for similarity transformations, noise, smoothing and cropping attacks.