Rotation invariant features: a method, problems and possible solutions

Dhabe, Priyadarshan; Bakki, S.R.; Kulkarni, Uday; Sontakke, T.R.

doi:10.1109/ijcnn.2002.1007759

Cited by 5 publications

(4 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These features are compared for Arabic OCR and a 100 % recognition rate with the NFD but in case of NCM the recognition rate reduces with change in image coordinates ratio is observed [2].…”

Section: Related Workmentioning

confidence: 99%

Fourier Features for the Recognition of Ancient Kannada Text

Soumya

Kumar

2015

Advances in Intelligent Systems and Computing

View full text Add to dashboard Cite

Optical Character Recognition (OCR) System for ancient epigraphs helps in understanding the past glory. The system designed here, takes a scanned image of Kannada epigraph as its input, which is preprocessed and segmented to obtain noise-free characters. Fourier features are extracted for the characters and used as the feature vectors for classification. The SVM, ANN, k-NN, Naive Bayes (NB) classifiers are trained with different instances of ancient Kannada characters of Ashoka and Hoysala period. Finally, OCR system is tested on epigraphical characters of 250 from Ashoka and 200 from Hoysala period. The prediction analysis of SVM, ANN, k-NN and NB classifiers is made using performance metrics such as Accuracy, Precision, Recall, and Specificity.

show abstract

Section: Related Workmentioning

confidence: 99%

Fourier Features for the Recognition of Ancient Kannada Text

Soumya

Kumar

2015

Advances in Intelligent Systems and Computing

View full text Add to dashboard Cite

show abstract

“…Let R = {rl,r2 ,..., rz}be a set of two-dimensional pattems then (p + qYh order regular geometrical moment [9] of all the patterns is given by, x=l y=l where the monomial x p y q gives displacement and the value of rh is taken as I , which represent force. Variables x and y are first and second components of two-dimensional pattern rh for k = 1,2 ,.,,.., z , and p, q are non-negative integers defining order of moment.…”

Section: Calculation Of Centroids Using Momentsmentioning

confidence: 99%

“…Then a pattem, with largest count along with its group of patterns giving membership grater than r is selected to create a cluster and its centroid is calculated using geometrical moments [9]. Thus a cluster is represented by a centroid and its outer boundary defined by parameter r , 'which controls total number of clusters created.…”

Section: Introductionmentioning

confidence: 99%

A new clustering method based on geometrical moment of patterns

Patil

Dhabe

Kulkarni

et al.

The 12th IEEE International Conference on Fuzzy Systems, 2003. FUZZ '03.

Self Cite

View full text Add to dashboard Cite

In this paper a new approach of moment based clustering is proposed, in which cluster centroids are calculated using regular geometrical moments of patterns. For this purpose zeroth order and first order moments are used. This approach is compared with Fuzzy Min-Max neural network (FMN) clustering algorithm. The performance of the proposed algorithm is found equiporable with FMN clustering, when tested for Fisher Iris data. The performance comparison is done at equal number of clusters created by both the algorithms by adjusting the network parameters. The recall of the proposed approach is found around three times faster than the FMN clustering algorithm.

show abstract

“…Mean point of the cluster is calculated by considering those patterns, which gives fuzzy membership greater than or equal to the centering parameter. Patil, Dhabe, Kulkarni and Sontakke [5] proposed moment based clustering approach (MBCNN) in which the centroid is calculated using geometrical moments [6]. Thus a cluster is represented by a centroid and its outer boundary defined by parameter r, which controls total number of clusters created.…”

Section: Introductionmentioning

confidence: 99%

A novel fuzzy clustering neural network

Patil¹,

Deshmukh²,

Mahajan³

Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005.

View full text Add to dashboard Cite

In this paper Fuzzy clustering Neural network (FCNN) is proposed with its learning algorithm, which utilizes fuzzy sets as cluster of patterns. The performance of FCNN is found better than FMN, FMPCNN, FHLSCNN and MBCNN clustering algorithms when compared with moderate number of clusters created. The cluster prototypes calculated reduces the confusion by giving fair treatment to the dense populated patterns. The total number of clusters created can be controlled by grouping factor A. The recall time per pattern of FCNN is smaller than the FMN, FMPCNN, FHLSCNN and MBCNN. Hence it can be used for real time applications.

show abstract

Rotation invariant features: a method, problems and possible solutions

Cited by 5 publications

References 4 publications

Fourier Features for the Recognition of Ancient Kannada Text

Fourier Features for the Recognition of Ancient Kannada Text

A new clustering method based on geometrical moment of patterns

A novel fuzzy clustering neural network

Contact Info

Product

Resources

About